A blade repair welding method fusing blue light scanning and welder sample learning

Abstract

The application relates to the technical field of blade repair welding, in particular to a blade repair welding method combining blue light scanning and welder sample learning. The method comprises the following steps: performing blue light scanning on a repair welding cavity to reconstruct a target curved surface and extract geometric and thermal boundary features, and establishing a local coordinate system; collecting and aligning trajectory, parameter and state data of skilled welders to construct a sample library; generating initial parameters based on feature matching samples, combining physical reference parameters and welder experience correction amounts; generating layered and channelized trajectories and controlling a robot to weld, adjusting parameters or sequences according to real-time state deviations; after completing the predetermined welding pass, performing scanning again, updating subsequent repair welding trajectories and repair welding parameters according to the difference between the actual surface and the target curved surface until completion. The application can realize digital migration of welder experience and full-process closed-loop control, suppresses layer error accumulation, and improves the consistency and adaptability of repair welding quality.

Description

Technical Field

[0001] This application relates to the field of blade repair welding technology, and in particular to a blade repair welding method that integrates blue light scanning and welder sample learning. Background Technology

[0002] Turbine blades, as core components of power equipment such as aero-engines and gas turbines, have complex structures with significant curvature variations and thin walls in some areas. During casting and long-term service, blades are highly susceptible to defects such as porosity, inclusions, cracks, or localized ablation. To avoid the huge economic losses caused by direct scrapping, repair is typically performed by removing defective structures followed by argon arc welding. Existing blade repair welding operations mainly rely on manual operation or basic robot teaching and reproduction modes. In manual operation, skilled welders adjust welding current, speed, wire feed, and welding torch posture in real time based on the geometry, location, and local heat dissipation conditions of the defects. In conventional automated solutions, welding tasks are often performed based on a preset 3D model that generates a fixed trajectory, supplemented by simple process monitoring methods.

[0003] However, existing welding repair technologies still face significant challenges in terms of adaptability and stability when dealing with irregular defects in blades, which are characterized by highly unstructured features. Because the location, shape, size, and surrounding thermal boundary conditions of defects vary greatly, fixed process procedures cannot cover all operating conditions, making it difficult to guarantee consistent welding quality. Furthermore, there is a complex coupling relationship between the heat accumulation state during welding, the dynamic behavior of the molten pool, and the final forming quality. Existing technologies often struggle to achieve accurate perception and closed-loop control of multi-physics states, easily leading to uncontrolled heat input, accumulation of interlayer errors, and process parameter mismatches due to the inability to quantify and transfer experience, thus limiting the widespread application of high-quality automated welding repair technologies. Summary of the Invention

[0004] This application provides a blade repair welding method that integrates blue light scanning and welder sample learning to solve the above-mentioned problems. The method includes: Blue light scanning was performed on the blade repair welding cavity after defect removal to reconstruct the target surface of the blade, extract the geometric features and thermal boundary features of the repair welding cavity, and establish a local coordinate system for the repair welding cavity; Collect the welding torch trajectory, welding process parameters, molten pool state and temperature state of skilled welders, convert the welding torch trajectory to the local coordinate system of the corresponding repair welding cavity, and align the welding process parameters, molten pool state and temperature state according to the repair welding filling progress. Combine the repair welding quality to build a welder repair welding sample library. Based on the geometric and thermal boundary characteristics of the cavity to be repaired, a welder's repair sample is matched. The physical reference parameters are determined based on the blade material, local wall thickness, and target weld size. Initial repair parameters are generated by using the parameter difference between the matched welder's repair sample and the corresponding physical reference parameters. Based on the repair welding cavity and the target curved surface, a layered and multi-pass repair welding trajectory is generated, the welding robot is controlled to perform repair welding, and the repair welding parameters or weld sequence is adjusted according to the deviation of the real-time molten pool state and temperature state relative to the corresponding state of the repair welding sample of the matched welder. After completing the predetermined weld bead, perform a blue light scan again, and update the subsequent weld trajectory and weld parameters based on the difference between the actual weld surface and the target surface until the weld is completed.

[0005] Optionally, the reconstructing of the target surface of the blade and the establishment of a local coordinate system for the weld repair cavity includes: The target surface is fitted with the scanning point cloud of the complete blade surface around the welding repair cavity. The centroid of the projection of the welding repair cavity on the target surface is taken as the origin of the coordinate system. The normal of the target surface at the centroid of the projection is taken as the height direction. The main extension direction of the projection contour of the welding repair cavity is taken as the first tangential direction. The local coordinate system of the welding repair cavity is established. The geometric features include the opening area, maximum depth, target filling volume, major-minor axis ratio, and boundary curvature of the welding cavity. The thermal boundary features include the minimum wall thickness, average wall thickness, shortest distance to the free edge of the blade, and shortest distance to the cooling hole in the region where the welding cavity is located.

[0006] Optionally, the welding fill progress is determined according to the following formula: ; in, Let t be the progress of the weld filler. The cumulative effective weld volume formed up to time t. The target filling volume for the corresponding weld repair cavity; The cumulative effective deposition volume is determined based on the cross-sectional area of ​​the welding wire, the wire feeding speed, the wire feeding time, and the pre-calibrated effective deposition coefficient, and the welding process parameters, molten pool state, and temperature state in different samples are matched according to the same repair welding filling progress.

[0007] Optionally, the matched welder repair sample includes: Samples that do not match the material type, welding wire material, or heat treatment status of the blade to be repaired are screened out from the welder's repair sample library; The remaining samples are standardized for target filling volume, maximum depth, aspect ratio, boundary curvature, minimum wall thickness, and shortest distance to cooling holes. The samples are then sorted according to the differences between each standardized feature and the corresponding features of the cavity to be repaired. Samples with qualified welding quality and feature differences not exceeding a preset difference threshold are selected from the ranking results as candidate samples. The sample weights are determined based on the feature differences and welding quality of the candidate samples. When the feature differences increase, the corresponding sample weight decreases. The sample weight of high-quality samples is greater than that of usable samples. The sample weights of each candidate sample are normalized so that the sum of the sample weights of each candidate sample is 1.

[0008] Optionally, the physical reference parameters include at least a reference welding current, a reference arc voltage, a reference welding speed, and a reference wire feed speed, and the physical reference parameters satisfy the following effective heat input per unit length: ; Within the process evaluation range corresponding to the blade material and local wall thickness, among which, Effective heat input per unit length For the effective thermal efficiency of the electric arc, Arc voltage For welding current, For welding speed; The initial welding parameters are determined according to the following formula: ; In the formula, These are the initial welding parameters corresponding to the filling progress. The physical reference parameters for the blade to be repaired are as follows: and The first The actual welding process parameters and their corresponding physical reference parameters for each candidate sample For the first The sample weights of each candidate sample.

[0009] Optionally, the generation of the layered and multi-pass welding trajectory includes: Based on the depth of the repair welding cavity and the preset single-layer deposition height, the repair welding cavity is divided into a bottom layer, at least one intermediate layer, and a surface layer; First, generate a contour weld bead extending along the boundary of the repair weld cavity at the bottom layer, and then generate an internal filler weld bead within the area enclosed by the contour weld bead; change the extension direction of the internal filler weld bead between adjacent intermediate layers; generate a shaping weld bead on the surface layer that is consistent with the local orientation of the target surface. The spacing between adjacent welds is determined based on the weld width and preset overlap rate in the candidate samples, and the spacing between adjacent welds is reduced at locations where the curvature of the repair cavity increases or the width decreases.

[0010] Optionally, the target state range of the blade to be repaired is determined based on the molten pool state, temperature state and sample weight of each candidate sample at the corresponding filling progress. The target state range includes the target ranges corresponding to the molten pool width, molten pool length, molten pool centroid offset, molten pool front temperature, molten pool rear temperature, and temperature difference between the two sides of the molten pool, wherein the first... Item status in fill progress The target value is determined according to the following formula: ; in, Let i be the target value of the i-th state. For the first The corresponding state values ​​of each candidate sample under the same filling progress. For the first The sample weights of each candidate sample; The upper and lower limits of the target state range are determined based on the degree of dispersion of the corresponding state values ​​of the candidate samples relative to the target value.

[0011] Optionally, the welding parameters or weld sequence can be adjusted according to the real-time molten pool condition and temperature condition, including: When the real-time weld pool width is higher than the upper limit of the corresponding target state range, reduce the welding current, increase the welding speed, or reduce the oscillation amplitude. When the real-time weld pool width is lower than the lower limit of the corresponding target state range and no continuous wetting area is formed on the side wall of the repair welding cavity, increase the welding current, reduce the welding speed, or extend the dwell time of the welding torch at the corresponding side wall. When the temperature behind the real-time molten pool is higher than the upper limit of the corresponding target state range, the subsequent welds adjacent to the current weld are stopped, and the next weld is selected from the unwelded welds whose corresponding area temperature is lower than the upper limit of the inter-pass temperature. If there are no unwelded welds whose corresponding area temperature is lower than the upper limit of the inter-pass temperature, the re-welding is suspended until the temperature of at least one unwelded weld area drops below the upper limit of the inter-pass temperature.

[0012] Optionally, the blue light scanning is performed again, including interlayer scanning of the weld repair area when a fill layer is completed, a preset number of weld beads are completed, or the actual weld bead height is detected to exceed the preset weld bead height range. The actual weld repair surface obtained by interlayer scanning is registered to the target surface. The area where the target surface is higher than the actual weld repair surface and the height difference is greater than the preset underfill threshold is determined as the remaining filling area. The number of subsequent filling layers, the number of weld beads, the weld bead spacing and the wire feed speed are updated according to the area, depth and local curvature of the remaining filling area. The area where the actual weld repair surface is higher than the target curved surface and the height difference is greater than the preset overfill threshold is defined as the overfill area. Subsequent weld beads are stopped in the overfill area, and the overfill area is reserved as the machining area after the weld repair.

[0013] Optionally, the cavity to be repaired is determined to be outside the coverage of the welder's repair sample library when at least one of the following conditions is met: the number of candidate samples is less than the preset number of samples; the number of candidate samples is greater than zero and the minimum feature difference among the candidate samples is greater than the preset difference threshold; the number of candidate samples is greater than zero and the local minimum wall thickness of the cavity to be repaired is lower than the minimum value among the local minimum wall thicknesses of each candidate sample. When the number of candidate samples is zero, automatic welding is stopped and a manual confirmation command is output; when the number of candidate samples is greater than zero and the cavity to be welded exceeds the coverage range, the welding robot is controlled to use the effective heat input per unit length to verify the parameters located in the lower half of the corresponding process evaluation range, and to perform verification welds at the bottom layer of the cavity to be welded, and to collect the molten pool state, temperature state and actual weld morphology of the verification welds. When the molten pool state and temperature state are both within the target state range determined based on candidate samples, and the actual weld bead morphology meets the preset weld bead width, height and continuity requirements, the subsequent repair welding parameters and repair welding trajectory are updated according to the actual parameters and actual weld bead morphology of the verified weld bead; otherwise, automatic repair welding is stopped and a manual confirmation command is output.

[0014] The above technical solution adopted in this application has the following advantages: By reconstructing the target surface of the blade using blue light scanning and extracting the geometric features and local thermal boundary features of the welding repair cavity, the welding repair decision can simultaneously consider the defect volume, shape, wall thickness, curvature, and surrounding heat dissipation conditions, thereby improving the identification accuracy of irregular blade defects and the targeting of the welding repair scheme.

[0015] By establishing a local coordinate system for the repair welding cavity and aligning the welding sample with the filling progress according to the ratio of the cumulative effective weld volume to the target filling volume, the influence of differences in blade clamping posture, defect location, and defect size on sample reuse is eliminated, and the experience of skilled welders can be effectively transferred between different repair welding tasks.

[0016] By linking and storing the quality results of the repair welding with the welding torch trajectory, welding process parameters, molten pool state and temperature state, the system learns the operating rules of skilled welders that have been verified by quality, thus avoiding interference from unqualified or unstable welding samples on automatic repair welding decisions.

[0017] By matching welder repair samples based on defect geometric features and thermal boundary features, and determining sample weights based on sample feature differences and repair quality, it is possible to select welder experience with higher applicability for the current repair cavity, reducing the adaptation error caused by repair welding only according to defect type or fixed parameter template.

[0018] By first determining the physical reference parameters that meet the requirements of blade material, local wall thickness, weld size, and heat input per unit length, and then superimposing the experience correction amount of skilled welders relative to the physical reference parameters, the learning model is prevented from directly outputting welding parameters lacking physical constraints, thereby improving the interpretability of the welding parameters and the safety of the process.

[0019] By generating layered and segmented welding trajectories and adjusting the welding sequence based on the real-time temperature of the welding area, continuous concentrated heating in thin-walled areas or adjacent areas can be avoided, reducing the risk of blade burn-through, thermal cracking, deformation, and expansion of the heat-affected zone.

[0020] By determining the molten pool state and temperature state of candidate welder samples under the corresponding filling progress as the dynamic target state range, and adjusting the welding current, welding speed, wire feed speed, welding torch posture and interpass waiting time according to the real-time state deviation, the robot repair welding process is transformed from execution of fixed parameters to closed-loop adjustment according to the molten pool formation and thermal state changes, thereby improving the weld fusion quality and forming stability.

[0021] By performing a blue light scan again after completing the predetermined weld or filler layer, the remaining filler area, subsequent trajectory, and weld parameters are updated based on the difference between the actual welded surface and the target surface. This can promptly eliminate the cumulative error between the actual weld deposit, weld width, and weld height and the offline prediction, thereby improving the final blade profile recovery accuracy.

[0022] By judging the coverage of the welder's weld repair sample library and performing low-heat input verification welds for defects that exceed the sample coverage, the robot can avoid directly performing complete automatic repair welding when there are no similar reliable samples, thus improving safety and fault tolerance under unknown defect conditions.

[0023] This application realizes the full-process recording of defect morphology before repair welding, welding process parameters, molten pool state, temperature change, trajectory adjustment, interlayer formation and final quality, transforming the traditional black-box repair welding process that relies on the welder's personal experience into an automated repair welding process that is learnable, monitorable, traceable and continuously optimized. Attached Figure Description

[0024] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0025] Figure 1 This is a flowchart illustrating a blade repair welding method that integrates blue light scanning and welder sample learning, as provided in an embodiment of this application. Detailed Implementation

[0026] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0027] Furthermore, the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. Additionally, the character " / " in this article, unless otherwise specified, generally indicates that the preceding and following related objects have an "or" relationship.

[0028] The embodiments of this application will now be described in further detail with reference to the accompanying drawings.

[0029] Example 1: like Figure 1 The diagram shown is a flowchart of a blade repair welding method integrating blue light scanning and welder sample learning provided in an embodiment of this application. The method includes: Step S101: Perform blue light scanning on the blade repair welding cavity after defect removal, reconstruct the target surface of the blade, extract the geometric features and thermal boundary features of the repair welding cavity, and establish a local coordinate system for the repair welding cavity.

[0030] This step aims to acquire precise 3D topography and heat dissipation environment information of the area to be repaired, providing geometric and physical benchmarks for subsequent sample matching and trajectory planning. First, a blue structured light scanning device is used to acquire high-precision 3D data of the repair cavity and the surrounding intact blade surface after defect removal processes such as grinding or milling. The scanning area not only covers the interior of the repair cavity but also extends to the intact surface ring outside the cavity boundary. Based on the acquired point cloud data, the target blade surface is reconstructed using point cloud fitting of the intact blade surface surrounding the repair cavity. This target surface characterizes the ideal blade surface morphology before the defect occurred and is used to determine the target contour after repair, the depth of the repair cavity, and the volume of space to be filled.

[0031] Based on the reconstructed target surface, the geometric and thermal boundary features of the welding repair cavity are further extracted. Geometric features characterize the spatial morphology of the welding repair cavity, specifically including the opening area, maximum depth, target filling volume, major-minor axis ratio, and boundary curvature. Thermal boundary features characterize the heat dissipation conditions of the welding repair area, specifically including the minimum wall thickness, average wall thickness, shortest distance to the blade's free edge, and shortest distance to the cooling holes in the region where the welding repair cavity is located. The local wall thickness can be determined based on the blade design model, bilateral blue light scanning results of the blade, ultrasonic thickness measurement results, industrial CT results, or a wall thickness database corresponding to known blade models.

[0032] To eliminate the impact of differences in blade clamping posture and defect location on data alignment, a local coordinate system for the welding repair cavity is established. Specifically, the centroid of the projection of the welding repair cavity onto the target surface is taken as the origin, the normal to the target surface at the centroid is taken as the height direction, and the main extension direction of the projected contour of the welding repair cavity is taken as the first tangential direction. Through this local coordinate system, subsequent data such as welding torch trajectory, scanning point cloud, and temperature detection position are uniformly transformed, achieving data standardization and transferability between welding repair tasks at different positions and postures.

[0033] Step S102: Collect the welding torch trajectory, welding process parameters, molten pool state and temperature state of skilled welders, convert the welding torch trajectory to the corresponding local coordinate system of the repair welding cavity, and align the welding process parameters, molten pool state and temperature state according to the repair welding filling progress, and construct a welder repair welding sample library based on the repair welding quality.

[0034] This step aims to transform the implicit operational experience of skilled welders into digital sample data that can be used by the robot. The data collection targets skilled welders with experience in repairing corresponding materials and blades, covering various defect volumes, opening shapes, defect depths, local curvatures, blade wall thicknesses, and material types. During manual repair welding, the system simultaneously collects multi-dimensional data: the position and posture of the welding torch tip (i.e., the welding torch trajectory), welding process parameters (including welding current, arc voltage, welding speed, wire feed speed, shielding gas flow rate, welding torch angle, oscillation amplitude, and dwell time), molten pool condition (images are acquired through a molten pool vision device, extracting features such as molten pool width, length, centroid, and wetting angle), and temperature conditions (temperatures in front of, behind, and to the sides of the molten pool).

[0035] To achieve effective alignment of welding cavity samples of different sizes and shapes, the acquired welding torch trajectory is converted to the corresponding local coordinate system of the welding cavity, and the welding process parameters, molten pool state, and temperature state are normalized according to the welding filling progress. The welding filling progress is defined. The cumulative effective weld volume formed up to time t Target filling volume of the corresponding repair welding cavity The ratio of the two values. The cumulative effective deposition volume is determined based on the wire cross-sectional area, wire feed speed, wire feed time, and a pre-calibrated effective deposition coefficient. By converting the above data into a function of the filling progress ξ, defect samples of different sizes can be compared and migrated under the same progress scale: initial filling – bottom layer completion – intermediate layer filling – surface shaping.

[0036] A welder's weld repair sample library was constructed based on the weld repair quality evaluation results. The weld repair quality evaluation items included the morphological deviation of the weld surface relative to the target surface, the presence of cracks or lack of fusion, the quantity and size of porosity and inclusions, the width of the heat-affected zone, the hardness change in the weld repair area, and the amount of blade deformation after weld repair. Samples in the library were categorized as high-quality, usable, and unqualified. Only samples with acceptable weld repair quality (e.g., high-quality or usable samples) were stored in the welder's weld repair sample library, while unqualified samples with cracks, burn-through, or obvious lack of fusion were removed to prevent the robot from learning incorrect operational experience.

[0037] Step S103: Match the welder's welding sample according to the geometric features and thermal boundary features of the cavity to be repaired, determine the physical reference parameters according to the blade material, local wall thickness and target weld size, and generate the initial welding parameters by using the parameter difference between the matched welder's welding sample and the corresponding physical reference parameters.

[0038] This step aims to generate initial process parameters for the current repair welding task that conform to both physical constraints and the experience of skilled welders. First, similar samples are selected from a welder's repair welding sample library based on the geometric and thermal boundary characteristics of the cavity to be repaired. The selection process includes: excluding samples whose material type, welding wire material, or heat treatment state is inconsistent with the blade to be repaired; standardizing the target filling volume, maximum depth, aspect ratio, boundary curvature, minimum wall thickness, and shortest distance to the cooling hole for the remaining samples; calculating and ranking the differences between each standardized feature and the corresponding feature of the cavity to be repaired; and selecting samples with acceptable repair welding quality and small feature differences as candidate samples. Sample weights are determined based on the feature differences and repair welding quality of the candidate samples, with weights decreasing as feature differences increase and increasing as the repair welding quality level improves.

[0039] After determining the matching sample, physical reference parameters are established based on the blade material, local wall thickness, and target weld bead size. These physical reference parameters include at least the reference welding current, reference arc voltage, reference welding speed, and reference wire feed speed. The determination of these parameters must meet the requirements for effective heat input per unit length. Within the process evaluation range corresponding to the blade material and local wall thickness (where η is the effective thermal efficiency of the arc, U is the arc voltage, I is the welding current, and v is the welding speed). For example, a lower upper limit of heat input is used in thin-walled areas, and a lower limit of heat input that can ensure sidewall fusion is used in the bottom contour establishment stage.

[0040] Subsequently, initial welding parameters are generated using the parameter differences between the matched welder's weld repair sample and the corresponding physical reference parameters. The calculation formula is as follows: In the formula, These are the initial welding parameters corresponding to the filling progress. The physical reference parameters for the blade to be repaired are as follows: and The first The actual welding process parameters and their corresponding physical reference parameters for each candidate sample For the first The sample weights of each candidate sample are determined. This method sets the initial welding parameters as a weighted sum of physical reference parameters and welder experience corrections, which ensures both the physical rationality of the parameters and inherits the fine-tuning experience of skilled welders for specific working conditions.

[0041] Step S104: Generate layered and multi-pass welding trajectories based on the welding cavity and target surface, control the welding robot to perform welding, and adjust the welding parameters or weld sequence based on the deviation of the real-time molten pool state and temperature state relative to the corresponding state of the matched welder's welding sample.

[0042] This step aims to plan a precise welding path and achieve closed-loop real-time control of the welding process. Based on the depth of the welding cavity and the preset single-layer deposition height, the welding cavity is divided into a bottom layer, at least one intermediate layer, and a surface layer. In the bottom layer, a contour weld bead extending along the boundary of the welding cavity is first generated, and then an internal filling weld bead is generated within the area enclosed by the contour weld bead. The extension direction of the internal filling weld bead is changed between adjacent intermediate layers. In the surface layer, a shaping weld bead consistent with the local orientation of the target surface is generated. During trajectory generation, the spacing between adjacent weld beads is determined based on the weld bead width and preset overlap rate in the candidate samples, and the spacing between adjacent weld beads is reduced at locations where the curvature of the welding cavity increases or the width decreases. After trajectory planning is completed, the welding robot is controlled to perform the welding operation according to the planned trajectory and initial welding parameters.

[0043] During the repair welding process, the system monitors the molten pool state and temperature in real time, and dynamically adjusts the repair welding parameters or weld sequence based on the deviations of these real-time states relative to the corresponding states of the welder's repair samples. Specifically, based on the molten pool state, temperature state, and sample weight of each candidate sample at the corresponding filling progress, the target state range of the blade to be repaired at the corresponding filling progress is determined. This target state range includes the target ranges corresponding to the molten pool width, molten pool length, molten pool centroid offset, temperature in front of the molten pool, temperature behind the molten pool, and temperature difference between the two sides of the molten pool.

[0044] When the real-time monitored state deviates from the target state range, corresponding adjustment strategies are implemented. For example, when the real-time weld pool width is higher than the upper limit of the corresponding target state range, the welding current is reduced, the welding speed is increased, or the oscillation amplitude is reduced; when the real-time weld pool width is lower than the lower limit of the corresponding target state range and a continuous wetting area is not formed on the sidewall of the repair welding cavity, the welding current is increased, the welding speed is reduced, or the dwell time of the welding torch at the corresponding sidewall is extended. When the temperature behind the real-time weld pool is higher than the upper limit of the corresponding target state range, the subsequent weld adjacent to the current weld is stopped, and the next weld is selected from the unwelded welds whose corresponding area temperature is lower than the upper limit of the inter-pass temperature; if there are no unwelded welds whose corresponding area temperature is lower than the upper limit of the inter-pass temperature, the repair welding is paused until the temperature of at least one unwelded weld area drops below the upper limit of the inter-pass temperature. Through this parameter and sequence adjustment based on real-time state feedback, it is ensured that the repair welding process is always within the reasonable process window defined by the high-quality welder sample.

[0045] Step S105: After completing the predetermined weld bead, perform blue light scanning again, and update the subsequent welding trajectory and welding parameters according to the difference between the actual weld surface and the target surface until the weld is completed.

[0046] This step aims to eliminate the cumulative error between offline planning and actual weld deposition through interlayer or staged rescanning, achieving precise closed-loop correction of the repair welding process. Interlayer scanning is performed on the repair area when a fill layer is completed, a preset number of weld passes are completed, or the actual weld pass height exceeds the preset range. The actual repair surface obtained from the interlayer scan is registered to the target surface, and the current repair weld formation quality is evaluated by comparing the difference between the actual repair surface and the target surface.

[0047] The area where the target surface is higher than the actual weldable surface and the height difference exceeds a preset underfill threshold is identified as the remaining fill area. The number of subsequent fill layers, weld beads, weld bead spacing, and wire feed speed are updated based on the area, depth, and local curvature of this remaining fill area. Simultaneously, the area where the actual weldable surface is higher than the target surface and the height difference exceeds a preset overfill threshold is identified as the overfill area. Subsequent weld bead generation is stopped in the overfill area, and this overfill area is retained as the machining area after the weld repair. Based on the updated trajectory and parameters, the welding robot continues to perform subsequent weld repair tasks, repeatedly executing the above scanning, evaluation, and updating process until the weldable area is completely filled and meets the preset completion conditions, thus completing the weld repair of the entire blade.

[0048] The aforementioned technical process acquires the precise geometric and thermal boundary features of the weld repair cavity through blue light scanning, and aligns them with sample data from skilled welders in the local coordinate system and filling progress dimension. This allows for the effective transfer of welding experience for defects of different locations and sizes. The introduction of physical reference parameters ensures the safety of the welding parameters within the process evaluation range, while the superposition of welder experience differences endows the parameters with adaptability and advancement for specific working conditions. During welding execution, dynamic adjustments are made based on the deviation of the real-time molten pool and temperature state from the target state of the sample, achieving precise control over complex heat accumulation and molten pool behavior. Finally, through interlayer blue light re-scanning, the subsequent trajectory and parameters are continuously corrected based on the difference between the actual weld repair surface and the target surface, effectively eliminating the cumulative error between offline planning and actual forming, ensuring the consistency and reliability of the final weld repair quality.

[0049] Example 2: In yet another embodiment, the process for reconstructing the target surface of the blade and establishing the local coordinate system of the weld repair cavity is provided for the embodiments of this application.

[0050] Step 1: Fit the target surface using the scanned point cloud of the complete blade surface around the weld repair cavity.

[0051] The scanned point cloud of the intact blade surface surrounding the weld repair cavity was selected as the base data for fitting the target surface. These point clouds were acquired by a blue light structured light scanning device from the weld repair cavity and its surrounding intact surface ring. Since the intact surface surrounding the weld repair cavity has not undergone defect removal processing, it retains the original aerodynamic profile characteristics of the blade. Using data from this region for fitting can reproduce the ideal surface morphology before the defect occurred to the greatest extent possible. By selecting an intact surface ring of a certain width outside the weld repair cavity boundary, mathematical fitting is performed based on quadratic surfaces, B-spline surfaces, or the original blade design model to generate a smooth and continuous target surface. This target surface not only serves as the geometric reference for subsequent weld repair but also determines the depth of the weld repair cavity and the target filling volume to be filled. By relying on the true morphology of the outer intact surface, the uncertainty caused by inference based solely on the cavity edge is overcome, significantly improving the reconstruction fidelity of the target surface.

[0052] Step 2: Use the centroid of the projection of the weld repair cavity onto the target curved surface as the origin of the coordinate system.

[0053] The opening contour of the welding repair cavity is vertically projected onto the target surface generated above, forming a projection area. The geometric centroid of this projection area, i.e., the center of mass or area of ​​the projection area, is calculated and set as the origin of the local coordinate system. The projection centroid has clear geometric representativeness and naturally reflects the spatial centroid position of the welding repair cavity on the target surface. This selection method ensures that no matter how the blade is clamped or rotated, the core position of the welding repair cavity is always anchored at the zero point of the local coordinate system, thus providing the necessary alignment reference for subsequent conversion of welding trajectories at different spatial positions to the same coordinate space, ensuring the translation invariance of the coordinate system.

[0054] Step 3: Take the normal of the target surface at the centroid of the projection as the height direction.

[0055] The normal vector of the target surface at the projection centroid is calculated, and its direction is set as the height direction (i.e., the Z-axis direction) of the local coordinate system. This direction is strictly determined based on the local geometric features of the target surface at the projection centroid, rather than selecting the global machining direction of the blade. Since the normal vector is always perpendicular to the tangent plane of the target surface, this definition method ensures that the height axis of the local coordinate system precisely matches the local surface undulations of the blade. During the repair welding process, the welding robot's welding torch typically needs to weld along the surface normal or at a certain angle. Aligning the height direction with the surface normal allows for direct mapping of the welding torch's attitude angle relative to the repair welding cavity. This setting eliminates the interference of the overall bending or twisting of the blade on the description of the welding torch trajectory, enabling accurate comparison and reproduction of repair welding operations in different curvature regions under a unified normal reference system.

[0056] Step 4: Take the main extension direction of the projected contour of the weld repair cavity as the first tangential direction and establish the local coordinate system of the weld repair cavity.

[0057] By analyzing the geometry of the projected profile of the welding repair cavity, the first tangential direction (i.e., the X-axis direction) in the horizontal plane is determined. This direction is typically selected from the major axis or principal extension direction of the projected profile, and is obtained through statistical calculations of the projection boundary point set using algorithms such as Principal Component Analysis (PCA). After establishing the first tangential direction, the second tangential direction (i.e., the Y-axis direction) is determined using the right-hand rule or combined with the natural orthogonal complement operation in the height direction, thus constructing a complete local three-dimensional Cartesian coordinate system for the welding repair cavity. This coordinate system has the projection centroid O as the origin, the target surface normal N as the Z-axis, and the projected profile principal extension direction L as the X-axis. The coordinate system established through this process achieves structural decoupling between the geometry and spatial attitude of the welding repair cavity. Based on this coordinate system, welding repair cavities of different positions and sizes can be mapped to a standardized description space, enabling the trajectory points, attitude angles, and process parameters in the welder's sample to be accurately transferred to the current welding repair task according to the spatial transformation relationship, achieving a precise correspondence between data and physical entities.

[0058] Geometric features include the opening area, maximum depth, target filling volume, major-minor axis ratio, and boundary curvature of the weld repair cavity.

[0059] A set of feature parameters that comprehensively characterize the geometry of the weld repair cavity are extracted from blue light scanning data to achieve accurate matching of weld repair samples. Among these, the opening area refers to the size of the projected area of ​​the weld repair cavity on the target surface, determining the coverage of a single weld pass or single layer of filler; the maximum depth refers to the farthest distance from the bottom of the weld repair cavity to the target surface, directly affecting welding heat input and layer planning strategies; the target fill volume refers to the spatial volume between the current surface of the weld repair cavity and the target surface, serving as a core basis for calculating wire consumption and fill progress; the major-minor axis ratio describes the flatness or elongation of the projected contour of the weld repair cavity, reflecting the extension characteristics of the defect; and the boundary curvature characterizes the degree of bending change at the edge of the weld repair cavity, providing important reference value for edge wetting and weld trajectory planning. These geometric features together constitute a multi-dimensional vector describing the spatial morphology of the weld repair cavity, providing a quantitative comparison basis for subsequent screening of welder samples with similar geometric conditions. Through refined geometric feature extraction, the welding characteristics of defects with different shapes can be distinguished, avoiding parameter misuse due to shape differences.

[0060] Thermal boundary characteristics include the minimum wall thickness, average wall thickness, shortest distance to the free edge of the blade, and shortest distance to the cooling hole in the area where the repair welding cavity is located.

[0061] Thermal boundary features closely related to heat conduction were extracted to assess heat accumulation and dissipation conditions during the repair welding process. Minimum and average wall thicknesses reflect the thickness of the metal structure in the repair area; thin-walled areas dissipate heat slowly and are prone to burn-through, requiring strict control of heat input. The shortest distance to the blade's free edge measures the distance of the repair location from the blade edge; edge areas typically dissipate heat quickly but are also prone to deformation. The shortest distance to the cooling holes considers the strong heat dissipation effect of the adjacent cooling structures; locations near cooling holes experience extremely rapid heat loss, potentially requiring increased heat input or changes in the welding sequence. These thermal boundary features were obtained through analysis of blade design models, bilateral scanning data, or thickness measurement data. Combining thermal boundary features with geometric features for sample matching ensures that selected welder samples are not only similar in shape but also comparable in terms of heat dissipation environment and thermophysical properties. This multi-dimensional feature fusion mechanism effectively solves the problem of incompatible process parameters caused by relying solely on geometric shape matching, improving the reliability of welding parameter transfer and the quality of the repair weld.

[0062] Example 3: In one optional implementation, a blade repair welding method integrating blue light scanning and welder sample learning, provided in this application embodiment, includes a process for determining the repair welding progress based on a formula. This method includes: Step 1: The progress of the filler weld is determined according to the following formula: , Let t be the progress of the weld filler. The cumulative effective weld volume formed up to time t. The target filling volume for the corresponding weld repair cavity.

[0063] Repair welding and filling progress This value characterizes the degree of completion of the current welding repair task over time, monotonically increasing from 0 in the initial state to 1 in the completed state as the welding progresses. Target fill volume. Defined as the total volume of weld material required to fill the space between the current surface of the repair cavity and the target surface to be reconstructed. This geometric parameter has been calculated in the blue light scanning and target surface reconstruction steps of the above embodiment. Cumulative effective weld volume. This reflects the total amount of effective welding material actually melted into the repair welding cavity up to time t. Through the above ratio calculation, the progress of the repair welding process is transformed from an absolute time scale affected by equipment differences and manual operation speed into a relative volume scale with clear physical meaning. This normalization process allows for horizontal comparison and status assessment of repair welding tasks with different defect sizes and welding speeds under a unified completion percentage benchmark, eliminating timing alignment deviations caused by workpiece geometric differences.

[0064] Step 2: The cumulative effective deposition volume is determined based on the cross-sectional area of ​​the welding wire, the wire feeding speed, the wire feeding time, and the pre-calibrated effective deposition coefficient.

[0065] Cumulative effective weld volume This is not a simple summation of wire feed rates, but rather the actual amount of metal deposited after accounting for process losses such as spatter, oxidation, and evaporation. The calculation of this parameter relies on a comprehensive consideration of the welding wire's geometric parameters and the process efficiency coefficient. During system initialization or process preparation, the diameter of the welding wire is first obtained, and the cross-sectional area of ​​the wire is calculated using the formula for the area of ​​a circle. Simultaneously, for specific combinations of welding process parameters (including welding current, arc voltage, shielding gas flow rate, etc.), the effective deposition coefficient κ is pre-calibrated through experiments. This coefficient characterizes the proportion of theoretical wire feed amount that is ultimately converted into effective deposited metal, and its value is usually less than 1, used to correct the deviation between the ideal wire feed amount and the actual deposition amount.

[0066] During the repair welding process, the instantaneous wire feeding speed is collected in real time by an encoder or sensor installed on the wire feeding mechanism. and cumulative wire feeding time. Based on the above parameters, the cumulative effective deposition volume can be expressed in integral form as: In the formula, The cross-sectional area of ​​the welding wire is constant; Let τ be the instantaneous wire feeding speed at time τ; κ be the pre-calibrated effective welding coefficient. By integrating the wire feeding speed over the time interval [0, t] and combining it with the cross-sectional area and the effective welding coefficient for weighted correction, the cumulative effective welding volume up to the current time t can be accurately calculated, thus providing accurate data support for progress calculation.

[0067] Step 3: And make corresponding welding process parameters, molten pool state and temperature state in different samples according to the same repair welding filling progress.

[0068] Given the significant differences in the volume of the repaired welding cavity corresponding to different skilled welder samples, aligning the data solely based on the absolute time axis would result in the inability to correctly map the features of small-volume defect samples to large-volume defect tasks, causing distortion in experience transfer. Therefore, the repaired welding filling progress ξ(t) defined in step one above is used as a unified time reference to normalize and align the multidimensional data in the welder repaired welding sample library.

[0069] In practice, welding process parameters (such as welding current, arc voltage, and wire feed speed), molten pool geometry (such as molten pool width, length, and wetting angle), and thermal field temperature conditions (such as temperature values ​​at specific locations before and after the molten pool) recorded in the sample library are all mapped to a functional relationship of the welding progress ξ. For example, when the progress of the current welding task reaches 30%, the system retrieves the parameter configurations and state characteristics corresponding to all historical samples at time ξ=0.3 from the welder sample library and uses them as the reference benchmark for the current stage. Through this alignment mechanism based on the filling progress, welding tasks of different scales can achieve precise feature matching in the same geometric filling stage (such as the bottom contour construction stage, intermediate filling stage, or capping stage), ensuring the accuracy and effectiveness of transferring welding expert experience under different working conditions.

[0070] Example 4: In another optional embodiment, the process of matching welder repair samples in a blade repair welding method that integrates blue light scanning and welder sample learning provided in this application embodiment is described. Matching welder repair samples includes: filtering out samples from the welder repair sample library that are inconsistent with the material type, welding wire material, or heat treatment state of the blade to be repaired.

[0071] Step 1: Perform strict filtering to quickly eliminate samples from the vast historical sample library that are absolutely unusable due to material or process incompatibility. The material type of the blade to be repaired (e.g., nickel-based superalloys, cobalt-based superalloys, titanium alloys, or stainless steel) directly determines the physical basis of the welding metallurgical behavior. If the base material corresponding to the sample in the sample library is different from the current blade, its welding thermal process, molten pool fluidity, and solidification characteristics will all be fundamentally different, and direct application will lead to repair welding failure. Similarly, the welding wire material (the welding material composition must match the base material) and heat treatment state (e.g., solution-treated or aged state) are also prerequisites for determining whether the welding process window is applicable. For example, for blades in an age-strengthened state, the allowable welding heat input is usually much lower than that in the solution-treated state. If the process parameters for a solution-treated sample are incorrectly matched, it is very easy to cause performance degradation in the heat-affected zone. Through the above strict screening based on material type, welding wire material, or heat treatment state, it can be ensured that subsequent matching work is only carried out in a sample space with similar physical characteristics, thereby avoiding uncontrollable risks caused by metallurgical incompatibility.

[0072] Step 2: Standardize the target filling volume, maximum depth, aspect ratio, boundary curvature, minimum wall thickness, and shortest distance to the cooling hole for the remaining samples. Sort the samples according to the differences between each standardized feature and the corresponding features of the cavity to be repaired.

[0073] The remaining samples selected are then subjected to quantitative comparison of multi-dimensional features. Since the six features—target filling volume, maximum depth, aspect ratio, boundary curvature, minimum wall thickness, and shortest distance to the cooling hole—have different physical dimensions and orders of magnitude (e.g., volume might be on the order of cubic millimeters, wall thickness on the order of millimeters, and curvature on the order of reciprocal millimeters), directly calculating the Euclidean distance would lead to large numerical features dominating the matching results, masking key differences (such as wall thickness) caused by small numerical features. Therefore, each feature is standardized. This standardization process can, for example, employ the Z-Score standardization method, mapping the corresponding features of the cavity to be repaired and the remaining samples to a dimensionless, uniform scale based on the statistical mean and standard deviation of each feature in the sample library, making each feature comparable in similarity calculations.

[0074] Based on standardization, the distance between the cavity to be repaired and each remaining sample in the aforementioned feature space is calculated. This distance measures the comprehensive similarity between the sample and the current task in terms of geometry (aspect ratio, boundary curvature, maximum depth, target filling volume) and thermal boundary conditions (minimum wall thickness, shortest distance to cooling holes). By calculating this distance, all remaining samples are ranked, and the ranking list reflects the global similarity ranking of the samples with the current repair task, providing an objective basis for subsequent optimization. Among them, the thermal boundary features such as minimum wall thickness and shortest distance to cooling holes are usually given high importance in the calculation, because these two directly determine the heat dissipation rate and heat accumulation sensitivity during the welding process, which are key constraints affecting the quality of repair welding.

[0075] Step 3: Select samples with qualified welding quality and feature differences not exceeding the preset difference threshold from the sorting results as candidate samples, and determine the sample weights based on the feature differences and welding quality of the candidate samples. When the feature differences increase, the corresponding sample weight decreases. The sample weight of high-quality samples is greater than the sample weight of usable samples. Normalize the sample weights of each candidate sample so that the sum of the sample weights of each candidate sample is 1.

[0076] Based on the aforementioned ranking results, a final set of candidate samples was further selected. The selection criteria had two constraints: first, the weld repair quality must be qualified, meaning the sample's quality inspection results after historical weld repairs (such as through X-ray inspection, ultrasonic testing, or metallographic testing) must meet preset acceptance standards, eliminating samples with cracks, incomplete fusion, or excessive porosity; second, the feature differences must not exceed a preset difference threshold. This dual selection process, using both quality and similarity thresholds, ensured that the candidate samples possessed good historical performance and were highly similar to the current task in terms of geometric and thermophysical properties.

[0077] After identifying candidate samples, sample weights are determined based on feature differences and weld repair quality levels. The calculation of these weights follows a clear physical logic: the smaller the feature difference of a sample (i.e., the closer it is to the previously ranked results), the higher the applicability of the welder's operational experience contained in that sample to the current defect, and therefore it should be assigned a greater weight; conversely, the sample weight decreases as feature differences increase. Simultaneously, a higher weld repair quality level (e.g., a high-quality sample is superior to a usable sample) indicates a more stable operational level and a higher yield rate, and its weight also increases accordingly. After determining the sample weights of each candidate sample, the sample weights are normalized so that the sum of the sample weights of all candidate samples is 1. This weight determination process can be formally expressed as: sample weights are positively correlated with weld repair quality levels and correlated with a negative exponential function of feature differences. Through this weighting mechanism, the most reliable process experience can be intelligently extracted from multiple similar samples, allowing high-quality and highly similar samples to play a dominant role in subsequent parameter generation, thereby significantly improving the robustness and success rate of robot weld repair parameter generation.

[0078] The above technical process, by constructing a rigorous sample screening and evaluation system, quantifies the implicit experience of skilled welders into calculable weighted indicators, ensuring that only those samples that are materially compatible, geometrically similar, and of excellent quality can participate in subsequent repair welding decisions, thus laying a solid data foundation for generating physically reasonable and technologically mature repair welding parameters.

[0079] Example 5: In yet another embodiment, this application also provides a process for determining physical reference parameters and initial welding parameters in a blade repair welding method that integrates blue light scanning and welder sample learning. In this method, the physical reference parameters include at least a reference welding current, a reference arc voltage, a reference welding speed, and a reference wire feed speed, and the physical reference parameters satisfy the following effective heat input per unit length: It falls within the process evaluation range corresponding to the blade material and local wall thickness, among which, Effective heat input per unit length For the effective thermal efficiency of the electric arc, Arc voltage For welding current, The welding speed is given; the initial welding parameters are determined according to the following formula: In the formula, These are the initial welding parameters corresponding to the filling progress. The physical reference parameters for the blade to be repaired are as follows: and These represent the actual welding process parameters and their corresponding physical reference parameters for the j-th candidate sample. Let be the sample weight of the j-th candidate sample.

[0080] Step 1: The physical reference parameters must include at least the reference welding current, reference arc voltage, reference welding speed, and reference wire feed speed. The physical reference parameters must satisfy the following requirements for effective heat input per unit length: It falls within the process evaluation range corresponding to the blade material and local wall thickness, among which, Effective heat input per unit length For the effective thermal efficiency of the electric arc, Arc voltage For welding current, This refers to the welding speed.

[0081] The process of determining the physical reference parameters is a constraint-based solution process constructed based on the inherent physical properties of the blades and the fundamental laws of welding heat conduction. This aims to provide a physical anchor point that meets the requirements of materials metallurgy and mechanics for subsequent transfer of welder experience. This mechanism effectively prevents defects such as burn-through caused by excessive heat input or incomplete fusion due to insufficient heat input caused by directly applying welder experience parameters.

[0082] First, based on the material type of the blade to be repaired (e.g., nickel-based superalloys, titanium alloys, etc.) and the local wall thickness, the corresponding process qualification range is queried or calculated. This process qualification range is determined through standard welding process qualification tests to ensure that the mechanical properties and non-destructive testing quality of the welded joint meet the requirements. On this basis, the effective heat input per unit length is used as the core constraint index to screen and determine the specific values ​​of the physical reference parameters. This process can be formally represented as the following decision function:

[0083] In the formula, This indicates the effective heat input per unit length, typically measured in J / mm. It represents the effective thermal efficiency of the electric arc and is a dimensionless coefficient whose value depends on the welding method (such as argon arc welding). For example, it can be set to a value between 0.6 and 0.8. This represents the arc voltage, measured in V. This indicates the welding current, expressed in amperes (A). This indicates the welding speed, expressed in mm / s.

[0084] against The value range is determined by differentiating the blade material and local wall thickness. For thin-walled areas with limited heat dissipation (such as areas with a wall thickness less than 70% of the standard wall thickness), Limit the heat level to a low level to avoid excessive heat accumulation that could lead to deformation or burn-through; for thick-walled areas with strong heat dissipation, appropriately increase the heat level. The upper limit of the allowable current is set to ensure sufficient penetration depth. Reference welding current. Reference arc voltage and reference welding speed The combination must strictly satisfy the above requirements. The interval constraint. Furthermore, the reference wire feed speed and welding speed... There is a matching relationship between the target weld cross-sectional area and the weld bead, which is usually calculated and determined based on the welding wire diameter and deposition efficiency to ensure the continuity of weld filling. The physical reference parameters generated by the above constraint solving mechanism ensure the stability of the welding thermal process at the physical level, providing safe and reliable reference operation instructions for robot repair welding.

[0085] Step 2: Determine the initial welding parameters according to the following formula: In the formula, These are the initial welding parameters corresponding to the filling progress. The physical reference parameters for the blade to be repaired are as follows: and These represent the actual welding process parameters and their corresponding physical reference parameters for the j-th candidate sample. Let be the sample weight of the j-th candidate sample.

[0086] The initial welding parameter generation logic adopts a composite decision-making architecture of physical benchmark and experience correction, explicitly decoupling the general physical laws of the welding process from the personalized implicit experience of skilled welders. This generation method differs from the existing approach of directly outputting parameters through a black box machine learning model. Instead, it obtains more universally applicable empirical knowledge by calculating the deviation of the welder's actual operation relative to their own physical conditions.

[0087] The parameter synthesis process can be formally represented as:

[0088] In the formula, This indicates the current progress of the welding repair task. The initial repair welding parameter vector is as follows; This indicates that the blades to be repaired, as determined in step one, are in progress. The physical reference parameter vector; Indicates the first Candidate samples (such as samples matched from a database of high-quality welders) are in progress. The actual welding process parameters are as follows; Indicates the first The physical reference parameters corresponding to each candidate sample are the theoretical reference values ​​derived from the blade material, wall thickness, and other conditions of the sample. Indicates the first The sample weights of each candidate sample reflect the similarity between the sample and the current task, as well as the quality of the weld repair. The weight values ​​can be, for example, normalized values, and satisfy the following conditions: .

[0089] In the formula The term represents the empirical correction or residual of a skilled welder relative to a theoretical physical reference under specific working conditions. For example, when dealing with certain curvature boundaries, a skilled welder might tend to use a slightly higher current than the physical reference to improve wetting; this fine-tuning experience is extracted as a residual. These empirical corrections from multiple similar samples are then superimposed onto the physical reference parameters for the current task using a weighted summation. This generation mechanism not only ensures that the initial filler welding parameters macroscopically conform to the process qualification range and heat input constraints (because the reference is physically safe), but also microscopically incorporates the welder's experience with different filler stages (…). The refined operation techniques significantly improved the rationality and interpretability of the welding parameters, providing high-quality input instructions for subsequent automatic robot execution.

[0090] The aforementioned technical process anchors basic parameters through physical constraints and incorporates expert experience using sample residuals, effectively solving the problems of poor parameter adaptability caused by simply relying on experience replication and the lack of physical meaning caused by simply relying on model prediction. By superimposing welder experience corrections while meeting the effective heat input constraint per unit length, this application ensures that the generated initial repair welding parameters possess both thermodynamic safety and retain the flexibility of manual operation, thereby enabling stable control of the molten pool morphology and heat-affected zone in actual welding and improving the consistency of repair welding quality.

[0091] Example 6: In another alternative embodiment, a flowchart of a layered and channeled welding trajectory generation method is provided for an embodiment of this application.

[0092] Step 1: Based on the depth of the repair welding cavity and the preset single-layer deposition height, divide the repair welding cavity into a bottom layer, at least one intermediate layer, and a surface layer.

[0093] This step aims to construct a macroscopic spatial layered architecture adapted to the needs of the repair welding process. The depth of the repair welding cavity can refer to the maximum distance from the surface of the repair welding cavity to the target curved surface obtained based on the aforementioned blue light scanning. This parameter determines the total thickness that needs to be filled. The preset single-layer deposition height is the ideal forming thickness of a single weld bead pre-set by the method provided in this application embodiment. Its value is not fixed, but is an engineering parameter determined comprehensively based on the welding wire diameter, the allowable heat input of the blade material, the local wall thickness, and the slope of the cavity boundary. The division process specifically involves dividing the total depth of the repair welding cavity by the preset single-layer deposition height to calculate the required sequence of layers. When there is a remainder that cannot be divided evenly, the deposition height of the top layer (i.e., the surface layer) is adjusted to ensure a smooth connection between the weld surface and the target curved surface. The bottom layer directly contacts the blade substrate, and its focus is on ensuring good fusion with the sidewall; the middle layer undertakes the main volume filling task; and the surface layer is related to the final geometric forming accuracy and aerodynamic performance. Through this layering, different layers undertake differentiated process functions, thereby achieving gradient control from the bonding strength of the bottom layer to the morphological accuracy of the surface layer.

[0094] Step 2: First, generate a contour weld bead extending along the boundary of the repair weld cavity on the bottom layer, and then generate an internal filler weld bead within the area enclosed by the contour weld bead.

[0095] To address the process requirement of ensuring sufficient metallurgical bonding between the welding material and the sidewalls and bottom of the welding cavity during bottom-layer repair welding to avoid incomplete fusion defects, a contour weld bead extending along the boundary of the welding cavity is first generated. The path of the contour weld bead is generated based on the contour line of the welding cavity boundary extracted by blue light scanning and offset calculation. When the welding torch executes this weld bead, the arc heat directly acts on the sidewall metal, achieving sufficient wetting and fusion of the sidewall. After the contour weld bead constructs a closed annular boundary, an internal filler weld bead is planned within the area enclosed by this boundary. The internal filler weld bead is used to quickly fill the large space inside the contour weld bead, and its trajectory can be generated using methods such as reciprocating straight lines, spirals, or equidistant offsets. This sequence of contouring first and then filling effectively utilizes the thermal restraint effect of the contour weld bead on the subsequent internal filler welding, reducing the risk of bottom cracking due to thermal shrinkage during the filling process.

[0096] Step 3: Change the extension direction of the internal filler weld between adjacent intermediate layers.

[0097] The intermediate layer, located between the bottom and top layers, typically occupies most of the filling volume of the weld repair cavity, and its welding heat input has the most significant impact on the heat accumulation of the blade. If the internal filler welds of the intermediate layer always maintain the same extension direction (e.g., all longitudinal), heat will be continuously input in a single direction, resulting in a large thermal stress gradient and uneven deformation inside the blade. Therefore, when generating the trajectory of adjacent intermediate layers, the extension direction of the internal filler welds is forcibly changed. For example, if the Nth intermediate layer uses a reciprocating weld along the X-axis, then the N+1th intermediate layer uses a reciprocating weld along the Y-axis, or a direction at a specific angle to the X-axis (e.g., 45° or 90°). This alternating change in interlayer direction allows the thermal cycle of the subsequent welding layer to partially offset or homogenize the stress field generated by the previous layer, avoiding unidirectional heat accumulation and localized overheating, thereby effectively controlling the amount of welding deformation of the blade.

[0098] Step 4: Generate a shaping weld bead on the surface that conforms to the local orientation of the target surface.

[0099] As the final working surface after welding repair, the geometry of the surface layer directly determines the blade's hydrodynamic performance; therefore, it is necessary to restore the original aerodynamic profile geometry of the blade. When planning the surface layer trajectory, the local geometric orientation of the target blade surface obtained from the aforementioned reconstruction is strictly followed. Specifically, the system calculates the streamline or isoparametric direction of the target surface within the welding cavity region, ensuring the trajectory of the shaping weld bead is consistent with these characteristic curves. The generation of the shaping weld bead no longer simply pursues filling efficiency, but prioritizes ensuring the flatness of the weld bead surface and the curvature continuity with the surrounding intact blade surface. By making the welding torch's movement path conform to the spatial posture of the target surface, the deposited metal naturally fills to the expected profile height, reducing subsequent machining allowances and ensuring a smooth transition between the welding repair area and the original blade surface.

[0100] Step 5: Determine the spacing between adjacent welds based on the weld width and preset overlap rate in the candidate samples, and reduce the spacing between adjacent welds at locations where the curvature or width of the repaired cavity increases.

[0101] The spacing between adjacent weld beads refers to the distance between the center lines of two parallel weld beads, which directly determines the overlap between weld beads. To achieve dense and uniform filling, the weld bead width recorded in the candidate samples matched in the above embodiment is first calculated based on a preset overlap rate. The preset overlap rate can be the ratio of the overlap width of two adjacent weld beads to the width of a single weld bead. This parameter is crucial for ensuring a smooth, groove-free weld surface and is usually determined by welding process qualification. The calculation logic for the reference spacing can be, for example, as: Spacing = Weld Bead Width × (1 - Overlap Rate). Based on this, a correction mechanism for complex curved surfaces is introduced: when an increase in the curvature of the repair cavity or a decrease in the width of the repair cavity is detected, the system automatically reduces the aforementioned spacing between adjacent weld beads. This is because in high-curvature or narrow-width regions, maintaining a large spacing can lead to excessively high or insufficiently covered weld beads on the outer edge, resulting in so-called overfill or underfill defects. By reducing the spacing and increasing the density of weld bead distribution, it is possible to adapt to the geometric shrinkage of curved surfaces, ensuring that the deposited metal can still cover these critical geometric transition areas evenly, and avoiding the formation of surface steps or voids.

[0102] The aforementioned technical process, through the synergistic effect of a layered architecture, differentiated weld bead strategies, and adaptive spacing adjustment, jointly solves the challenges of forming quality and stress control in the welding of complex curved blade surfaces. The macroscopic layering establishes the functional positioning of each layer; the contour and filling strategy of the bottom layer ensures bonding strength; the directional design of the middle layer disperses thermal stress; and the shaping weld bead of the surface layer locks in geometric accuracy. The microscopic adaptive spacing adjustment ensures the uniform spread of the filler material under different curvature and width conditions. Through this multi-level, adaptive trajectory planning, the welding process can meet the stringent requirements of the blade's aerodynamic profile while effectively controlling welding deformation and defects, achieving an automated welding closed loop from geometric mapping to precise execution.

[0103] Example 7: In one possible implementation, this application also provides a method for determining the target state range of the blade to be repaired under the corresponding filling progress based on the molten pool state, temperature state, and sample weight of each candidate sample under the corresponding filling progress.

[0104] Step 1: Based on the molten pool state, temperature state, and sample weight of each candidate sample at the corresponding filling progress, determine the target state range of the blade to be repaired at the corresponding filling progress.

[0105] To achieve precise closed-loop control of welding quality, dynamic and quantifiable control objectives need to be established. This step aims to transform the implicit experience contained in the multiple high-quality welder repair samples selected in the previous steps into specific state monitoring indicators that can be executed by the machine. Unlike existing technologies that only imitate process parameters such as current and voltage, this embodiment delves into the physical reaction level during welder operation, namely the evolution of the molten pool morphology and the distribution of the thermal field, using this as the control benchmark.

[0106] The aforementioned state data originates from candidate samples that have passed quality evaluation in the welder repair sample library. During the sample acquisition phase, a weld pool vision device (such as a high-speed industrial camera) and an infrared thermal imager, mounted on the end of the welding torch, are used to simultaneously record the weld pool image sequence and temperature field data during the repair welding process of skilled welders. Through image processing and algorithm analysis, key physical quantities, namely the aforementioned defined weld pool states and temperature states, are extracted from the raw data. Given the differences in the size of different repair welding cavities and the required welding time, all state data are strictly aligned with the time axis according to the aforementioned defined repair welding filling progress ξ, thereby ensuring comparability between different samples.

[0107] Based on the target state range constructed through the above processing, a clear state benchmark is provided for robot welding repair. When the real-time monitored molten pool state and temperature state fall within this range, it indicates that the current welding repair process meets the operating level of a skilled welder; if it deviates from this range, a parameter adjustment mechanism is triggered. This control method effectively overcomes the problems of existing technologies that rely solely on fixed process parameters and are difficult to cope with fluctuations in actual working conditions such as sudden changes in heat dissipation conditions or gap variations.

[0108] Step 2: The target state range includes the target ranges corresponding to the molten pool width, molten pool length, molten pool centroid offset, molten pool front temperature, molten pool rear temperature, and temperature difference between the two sides of the molten pool.

[0109] The target state range is a multi-dimensional set encompassing key indicators of molten pool geometry and thermodynamic characteristics. Specifically, this range includes target intervals across the following six core dimensions: The weld pool width refers to the maximum lateral dimension of the molten metal covering the surface of the base material in a plane perpendicular to the welding direction. This feature reflects the lateral spread of arc energy on the base material surface and the fusion of the sidewalls. The weld pool length refers to the maximum distance along the welding direction from the arc cathode spot (or the point of action of the welding wire end) to the tail of the weld pool. This feature characterizes the heat penetration depth of the heat source in the direction of movement and the dragging shape of the weld pool. The weld pool centroid offset refers to the spatial distance deviation of the geometric center (centroid) of the weld pool area relative to the preset welding torch trajectory centerline or planned path. This feature is used to measure the welding torch alignment accuracy, the effect of magnetic deflection, and the symmetry of the molten metal flow.

[0110] The temperature in front of the molten pool refers to the surface temperature of the unmelted but heat-affected area of ​​the base material (front end of the heat-affected zone) located at the forefront of the molten pool. This characteristic reflects the preheating effect and the rate of heat transfer to the unwelded area. The temperature behind the molten pool refers to the surface temperature of the solidified or solidifying weld and heat-affected zone located at the tail end of the molten pool. This characteristic is directly related to the cooling rate, solidification microstructure, and residual stress level. The temperature difference between the two sides of the molten pool refers to the temperature difference between the left and right edge regions of the molten pool in a cross-section perpendicular to the welding direction. This characteristic reveals the non-uniformity of heat distribution in the lateral direction and is important for preventing asymmetric shrinkage and cracking.

[0111] The aforementioned six dimensions of state data collectively constitute a complete portrait of the welding process. By defining the target range corresponding to each of these six indicators, this application can delve from macroscopic heat input control to microscopic molten pool dynamics control, ensuring that the robot can perform precise operations based on the real-time dynamics of the molten pool, achieving a control effect similar to that of a skilled welder.

[0112] Step 3: Wherein, the state of the i-th item is in the filling progress The target value is determined according to the following formula:

[0113] in, Let i be the target value of the i-th state. Let be the corresponding state value of the j-th candidate sample under the same filling progress. Let be the sample weight of the j-th candidate sample.

[0114] After defining the constituent dimensions of the target state range, it is necessary to calculate the position of each dimension at a specific weld filling progress. The specific target value is defined below. This target value represents the ideal state level that the current repair welding task should achieve at a certain moment or in a certain filling stage.

[0115] Given that there are usually multiple matched candidate samples (let's say n), and each sample has a different degree of similarity to the current task (determined by sample weights), (Characteristics), this embodiment uses a weighted fusion method to determine the final target value. For any of the above six states... The status of the item (e.g., weld pool width) is such that the filler welding progress is... At that time, its target value The following formula is used to calculate:

[0116] In the formula, Indicates the first Each candidate sample is at the same filling progress. Next, the Historical measured values ​​of the item's state. For example, when When representing the width of the molten pool, That is, the first Each sample was completed The filler ratio is the width of the weld pool formed by a skilled welder. For the first The sample weight of each candidate sample is calculated in the sample matching step based on the similarity between the sample and the current weld repair cavity in terms of geometric and thermal boundary features, as well as the weld repair quality level of the sample itself.

[0117] Through the above formula, this embodiment can integrate the experience and wisdom of multiple skilled welders. If a sample is highly similar to the current task (i.e. The larger the similarity (the greater the similarity), the more dominant the corresponding state value in the target value; conversely, samples with lower similarity have a smaller impact. This weighted averaging mechanism not only smooths out the random bias that may exist in a single sample, but also adaptively integrates the operating styles of different welders, making the generated target value both physically based and broadly representative.

[0118] Step 4: Determine the upper and lower limits of the target state range based on the dispersion of the corresponding state values ​​of the candidate samples relative to the target value.

[0119] target value A central benchmark for condition control was established. However, due to factors such as material inhomogeneity and power grid fluctuations, the condition values ​​inevitably fluctuate during the actual welding process. Therefore, a reasonable allowable fluctuation range needs to be set around the target value, i.e., the upper and lower limits of the target condition range. The core of this step lies in using the statistical characteristics of candidate sample data to determine the width of this range.

[0120] The degree of dispersion reflects the progress of each candidate sample at the same filling rate. Next, for the first The consistency of the state performance. If the dispersion is small, it means that all high-quality samples perform very similarly in this state, which means that this state is extremely sensitive to welding quality and should be strictly controlled; if the dispersion is large, it means that there are large differences in operation among different skilled welders in this state, which may mean that this state is greatly affected by the working conditions, or that a wider adjustment range is allowed.

[0121] In this embodiment, the upper limit of the target state range and lower limit Based on the candidate sample state value Relative to the weighted target value The upper limit is determined by the statistical dispersion (such as standard deviation, variance, or interquartile range). For example, the upper limit can be set as the target value plus... The standard deviation is set to times the target value, with the lower limit set at the target value minus [missing information]. The standard deviation is times the coefficient. It can be flexibly set according to the requirements of control accuracy (e.g., 1.5 or 2) to cover the distribution range of most high-quality samples.

[0122] By determining the upper and lower limits based on the degree of dispersion, this embodiment can generate adaptive control thresholds. For states with high consistency, the system automatically generates a narrower tolerance band, prompting the robot to perform precise control; for states with large differences, the system generates a wider tolerance band, avoiding frequent and unnecessary parameter adjustments due to overly strict constraints. This dynamic target state range mechanism significantly improves the adaptability and robustness of the welding repair process to complex and changing actual working conditions.

[0123] In summary, this embodiment establishes a closed-loop control benchmark based on multi-dimensional state monitoring and dynamic weight fusion. The weld pool width and length ensure reasonable metal spreading; the weld pool centroid offset guarantees the accuracy of the weld trajectory; the temperatures in front of and behind the weld pool control the thermal cycle and cooling rate; and the temperature difference on both sides of the weld pool maintains the symmetry of the thermal field. The synergistic effect of these six dimensions enables the robotic welding process to not only meet macroscopic geometric requirements but also approach the level of a skilled welder in terms of microstructure formation and stress control, ultimately ensuring the stability and reliability of the weld quality.

[0124] Example 8: One possible implementation is a method flow provided in this application embodiment for adjusting the welding parameters or weld sequence based on the real-time molten pool state and temperature state.

[0125] Step 1: When the real-time weld pool width is higher than the upper limit of the corresponding target state range, reduce the welding current, increase the welding speed, or reduce the oscillation amplitude.

[0126] During the repair welding process, a visual sensing device is used to monitor the geometry of the molten pool in real time, and the molten pool width is extracted as a key feature parameter. If the monitored real-time molten pool width exceeds the upper limit of the target state range determined based on candidate samples, it indicates that the current heat input level is too high or the molten pool stability is insufficient, posing a risk of excessive excess weld height, undercut, or uncontrolled flow of deposited metal. In this case, a parameter correction strategy is implemented to reduce the molten pool size.

[0127] Welding current is a core process parameter that determines the arc heat input and molten pool volume. Reducing the welding current directly decreases the arc's heat power and penetration, causing an excessively large molten pool to shrink in size. Welding speed determines the residence time of the heat source on the surface of the welding cavity; increasing the welding speed reduces the heat input per unit length, thus limiting the lateral expansion of the molten pool. The oscillation amplitude reflects the lateral oscillation range of the welding torch perpendicular to the welding direction; reducing the oscillation amplitude concentrates the heat source energy and reduces the spread of the molten pool in the width direction. These three adjustment methods can be implemented individually or in combination. For example, one can first try reducing the welding current; if the response speed is insufficient, the welding speed can be increased simultaneously. Alternatively, when using a narrow gap welding strategy, the oscillation amplitude can be reduced first to avoid touching the cavity sidewall. This tiered adjustment mechanism ensures that when dealing with the abnormal state of an excessively wide molten pool, the molten pool shape can be quickly pulled back to the target range through a physically constrained combination of parameters, avoiding arc instability or weld formation defects caused by large fluctuations in a single parameter.

[0128] Step 2: When the real-time weld pool width is lower than the lower limit of the corresponding target state range and no continuous wetting area is formed on the side wall of the repair welding cavity, increase the welding current, reduce the welding speed, or extend the dwell time of the welding torch at the corresponding side wall.

[0129] When the width of the molten pool monitored in real time is lower than the lower limit of the target range, it usually means that the heat input is insufficient. If a continuous wetting area has not yet formed on the sidewall of the repair welding cavity at this time, quality problems such as incomplete fusion or poor sidewall connection are very likely to occur. Therefore, when it is determined that the molten pool is too narrow and the wetting is insufficient, a control strategy that increases the heat input and improves the fluidity of the molten pool should be implemented.

[0130] Increasing the welding current increases the arc blowing force and molten pool temperature, enhancing the fluidity of the molten metal and its penetration into the sidewalls, thus helping to widen the molten pool and fill gaps in the sidewalls. Decreasing the welding speed prolongs the heating time at a specific location, allowing the molten metal sufficient time to spread to both sides, especially in areas where wetting is difficult on the sidewalls. Extending the heat treatment time is crucial for eliminating incomplete fusion. The welding torch dwell time refers to the instantaneous duration the torch is held at the extreme position on the sidewall during oscillating welding. Extending the dwell time at the corresponding sidewall utilizes the cathode atomization and direct heat conduction of the arc to forcibly increase the local temperature of the sidewall, promoting thorough wetting and fusion of the molten metal with the base material. For example, for repair cavities with a large depth-to-width ratio, the dwell time when oscillating to the sidewall can be extended from the conventional 0.1 s to 0.3 s or longer, and the welding speed can be appropriately reduced to ensure complete fusion at the root of the sidewall. This step, through coordinated adjustment of current, speed, and timing, not only ensures that the molten pool size meets the requirements, but also addresses the technical challenge of poor sidewall wetting, thereby improving the connection reliability of the repair weld joint.

[0131] Step 3: When the temperature behind the real-time molten pool is higher than the upper limit of the corresponding target state range, stop executing the subsequent welds adjacent to the current weld and select the next weld from the unwelded welds whose temperature in the corresponding area is lower than the upper limit of the inter-pass temperature.

[0132] During blade repair welding, heat accumulation control is crucial to prevent deformation and cracking. When the temperature monitoring device detects that the temperature behind the molten pool exceeds the upper limit of the corresponding target state range, it indicates that the heat load in that area has approached or exceeded the material's thermal damage threshold. At this point, a more aggressive weld sequence scheduling strategy should be adopted.

[0133] First, immediately stop welding any subsequent welds adjacent to the current high-temperature weld. This is because continuous welding of adjacent welds would further exacerbate heat accumulation in the area, leading to coarsening of the heat-affected zone or liquefaction cracks. Then, scan the temperature status of all unwelded welds, selecting candidate welds whose current temperature is below the preset inter-pass temperature limit, and choosing the next weld from these candidates. This skip-welding strategy fully utilizes the self-heating capacity of the unwelded areas of the blade, allowing cooling time for the high-temperature areas by dispersing heat sources. For example, when the temperature on the left side of the repair welding cavity reaches the inter-pass temperature limit, it can automatically switch to the lower-temperature unwelded welds on the right or bottom side to continue work, instead of continuing to stack welds on the left. Through this dynamic weld scheduling, without interrupting the overall workflow, the peak temperature and temperature gradient of the repair welding area are effectively reduced, and the temperature field distribution of the blade is optimized, thereby significantly reducing the risk of hot cracking and welding deformation.

[0134] Step 4: If there are no unwelded welds with a corresponding area temperature lower than the upper limit of the interpass temperature, suspend the re-welding until the temperature of at least one unwelded weld area drops below the upper limit of the interpass temperature.

[0135] The above-mentioned weld sequence adjustment strategy is based on the premise that there are unwelded welds that meet the temperature requirements. When all unwelded welds have been traversed and no area is found with a temperature lower than the upper limit of the inter-pass temperature, or when all candidate welds are in a high-temperature state, it means that the current repair welding task has entered the heat saturation stage. At this time, continuing welding will inevitably lead to quality deterioration. Therefore, the forced cooling mechanism is triggered, and the repair welding operation is suspended.

[0136] During the pause, the welding robot stops igniting the arc, and the protective gas is delayed to prevent high-temperature oxidation of the molten pool. The blade to be welded is cooled by natural convection or auxiliary air cooling while stationary. The system continuously monitors the real-time temperature changes of each unwelded area until the temperature of at least one area drops below the upper limit of the interpass temperature. Once this restart condition is met, the pause is automatically lifted, and the system returns to step three to select the next weld pass from the cooled area to continue welding. This safety fallback mechanism ensures that the welding process will not exceed the thermophysical performance limits of the material under any extreme operating conditions, thereby preventing blade overheating, degradation of microstructure, or burn-through accidents caused by blind continuous operation, and providing a safety barrier for the welding process.

[0137] Through the above steps, the method provided in this application constructs a hierarchical closed-loop control system of parameter fine-tuning, weld bead scheduling, and forced cooling. This system can not only precisely adjust microscopic process parameters such as current, speed, and swing amplitude based on the real-time deviation of the weld pool width to solve the weld pool forming problem; it can also macroscopically optimize the weld bead execution sequence or forcibly interrupt the operation based on the overall temperature state of the repair welding area to solve the heat accumulation control problem. This multi-scale, multi-level collaborative control logic enables the robot to, like a skilled welder, pursue welding efficiency while considering the thermal sensitivity of the blades, thereby stably outputting welds that meet quality requirements in the complex and ever-changing repair welding cavity environment.

[0138] Example 9: In another optional embodiment, a process for performing blue light scanning again and updating subsequent weld repair trajectories provided in this application embodiment further includes: performing interlayer scanning on the weld repair area when a fill layer is completed, a preset number of weld beads are completed, or the actual weld bead height is detected to exceed a preset weld bead height range; registering the actual weld repair surface obtained from the interlayer scanning to the target surface; determining the area where the target surface is higher than the actual weld repair surface and the height difference is greater than a preset underfill threshold as the remaining fill area; updating the number of subsequent fill layers, the number of weld beads, the weld bead spacing, and the wire feed speed according to the area, depth, and local curvature of the remaining fill area; determining the area where the actual weld repair surface is higher than the target surface and the height difference is greater than a preset overfill threshold as the overfill area; stopping the generation of subsequent weld beads in the overfill area; and retaining the overfill area as the machining area after weld repair.

[0139] Step 1: When a fill layer is completed, a preset number of weld beads are completed, or the actual weld bead height is detected to exceed the preset weld bead height range, perform an interlayer scan on the weld repair area.

[0140] By introducing an interlayer scanning mechanism, closed-loop monitoring and error correction of the welding process are achieved. The interlayer scanning trigger employs a multi-condition redundancy design to eliminate accumulated errors arising from purely theoretical predictions. When a complete fill layer is finished during the repair welding operation, indicating that all weld beads in that layer have been deposited, interlayer scanning can be performed to assess the flatness and allowance distribution of the entire layer. For deeper repair cavities, to avoid difficulties in repair due to deviations discovered after a single layer is completed, a preset number of weld beads is set as the trigger condition. This preset number can be, for example, 2 to 5 beads, depending on the cavity's geometry and thermal sensitivity. Furthermore, when the visual monitoring system detects that the actual weld bead height exceeds the preset weld bead height range (e.g., the actual weld bead height is higher than the upper limit of this range), interlayer scanning is immediately triggered to prevent further expansion of defects. Before performing interlayer scanning, it is necessary to confirm that the temperature of the repair area has dropped to within the allowable operating temperature range of the blue light scanning equipment, for example, below 50°C, to avoid high temperatures affecting imaging accuracy or damaging the equipment. Subsequently, a blue light structured light scanning device was used to collect three-dimensional point cloud data of the weld repair area, obtaining high-precision data including the surface morphology of the welded weld and the bottom morphology of the unfilled area.

[0141] Step 2: Register the actual weld repair surface obtained from the interlayer scan to the target surface. The area where the target surface is higher than the actual weld repair surface and the height difference is greater than the preset underfill threshold is determined as the remaining fill area. Then, update the number of subsequent fill layers, the number of weld beads, the weld bead spacing, and the wire feed speed according to the area, depth, and local curvature of the remaining fill area.

[0142] After acquiring the interlayer scan data, the point cloud of the actual weld repair surface is precisely registered with the target surface obtained from the aforementioned reconstruction. The registration process uses the unchanged intact blade surface ring around the weld repair cavity as the reference region. Through the Iterative Closest Point (ICP) algorithm or other fine registration algorithms, the point cloud of the actual weld repair surface in the interlayer scan coordinate system is transformed to the coordinate system of the target surface. After registration, the height difference between the target surface and the actual weld repair surface is calculated to identify underfilled and overfilled areas. The height of the target surface is denoted as... The actual height of the repaired weld surface is recorded as For any point When satisfied And height difference When the underfill threshold is greater than the preset underfill threshold, the point is determined to belong to the remaining filling area. The preset underfill threshold is set according to the metallurgical quality requirements of the blade material and the allowance allocation strategy for subsequent machining. For example, it can be set to 0.1 mm to 0.3 mm to filter out small surface roughness fluctuations and avoid unnecessary updates to the welding repair plan due to measurement noise.

[0143] After determining the remaining filler area, the subsequent welding scheme is dynamically replanned based on the geometric features of this area. First, the geometric features of the remaining filler area are extracted, including area, maximum depth, and local curvature. Based on the ratio of the maximum depth of the remaining filler area to the preset single-layer deposition height, the required number of subsequent filler layers is recalculated and updated. For example, if the remaining depth is insufficient for a single-layer deposition height, only one surface filler layer needs to be planned. Based on the area and contour shape of the remaining filler area, the number of subsequent weld passes is updated. For cases with significant area reduction, the number of internal filler passes is reduced accordingly, or the filler strategy is adjusted. Local curvature is obtained by analyzing the surface normal variation rate of the bottom and sidewalls of the remaining filler area. Areas with high curvature usually indicate increased difficulty in adjusting the welding torch posture and significant influence of the molten pool flow on the sidewalls. For high-curvature areas, the coverage accuracy is improved by reducing the spacing between adjacent weld passes, for example, reducing the standard spacing by 10%–20%. Simultaneously, to prevent excessive heat input accumulation due to reduced spacing, the wire feed speed is appropriately reduced. Conversely, for areas with gentle curvature and shallow depth, the weld pass spacing can be appropriately increased to improve welding efficiency. The decision to update the wire feeding speed takes into account both the correction amount of the remaining filling volume and the heat input constraint. When the remaining volume is less than the original planned value, the wire feeding speed is reduced accordingly to match the actual filling requirements, thereby avoiding material waste and excessive margin.

[0144] Step 3: The area where the actual weld repair surface is higher than the target curved surface and the height difference is greater than the preset overfill threshold is identified as the overfill area. The generation of subsequent weld beads in the overfill area is stopped, and the overfill area is retained as the machining area after the weld repair.

[0145] During the registration and height difference calculation process, for those that satisfy Points with a height difference greater than the preset overfill threshold are identified as overfilled areas. Overfilling may occur due to actual deposition efficiency exceeding prediction or interlayer deformation. Once an overfilled area is identified, an exclusion operation is performed in subsequent weld path planning. This means stopping the generation of any new weld paths within the overfilled area and prohibiting the robot from continuing welding in that area to prevent further overfilling leading to stress concentration or increased repair costs. This overfilled area is not considered a defect but is intentionally reserved for machining after the weld repair. Based on the final aerodynamic profile requirements of the blade and machining capabilities, the overfilled area is marked as the material portion that needs to be removed through subtractive machining methods such as milling and grinding. This approach converts positive dimensional deviations during the weld repair process into machining allowances, utilizing the high precision of machining to ensure the geometric consistency of the final blade profile while avoiding the increased difficulty in welding control and reduced weld repair efficiency caused by pursuing zero overfill.

[0146] The above steps work together to obtain the actual weld repair morphology through interlayer scanning, and combine it with the target surface for precise comparison and diagnosis, thereby driving the dynamic update of trajectory and parameters. At the same time, overfill and underfill areas are properly handled, thereby eliminating the accumulation of deviations between offline planning and the actual welding process, and ensuring that the final quality of the weld repair operation meets the design requirements.

[0147] Example 10: In another optional embodiment, the method provided in this application further includes a process of determining that the cavity to be repaired exceeds the coverage range and performing a verification weld. This process is used to establish a risk control mechanism during the automatic repair welding process to ensure that when the characteristics of the cavity to be repaired exceed the reliable coverage range of the existing welder repair sample library, the system can safely handle unknown working conditions and avoid welding defects or workpiece damage due to blindly migrating mismatched experience.

[0148] Step 1: When at least one of the following conditions is met, the cavity to be repaired is determined to be outside the coverage of the welder's repair sample library: the number of candidate samples is less than the preset number of samples; the number of candidate samples is greater than zero and the minimum feature difference among the candidate samples is greater than the preset difference threshold; the number of candidate samples is greater than zero and the local minimum wall thickness of the cavity to be repaired is lower than the minimum value among the local minimum wall thicknesses of each candidate sample.

[0149] A multi-dimensional boundary check is performed to verify the matching degree between the cavity to be repaired and the current welder's repair sample library, in order to identify whether the sample library has the ability to support the current repair task. This judgment process is based on three core dimensions: sample quantity, feature similarity, and physical structure safety. Specifically, if the number of candidate samples is less than the preset sample quantity, it indicates a lack of effective data resources in the sample library similar to the current repair task, failing to provide statistically reliable parameter references. The preset sample quantity can be determined empirically, such as setting it to 5 or 10, to ensure that the sample set has a certain representativeness. If the minimum feature difference is greater than the preset difference threshold, it indicates that the selected optimal sample still has significant deviations from the cavity to be repaired in key features such as geometry and heat dissipation conditions. Directly using its parameters may lead to unstable repair quality. The minimum feature difference can usually refer to the Euclidean distance or Mahalanobis distance between the cavity to be repaired and the candidate sample in the standardized feature space. The preset difference threshold is set according to the allowable feature deviation range in historical successful repair cases, for example, it can be set to 0.2 or 0.3 (normalized values). If the minimum local wall thickness of the cavity to be repaired is lower than the minimum wall thickness of the corresponding candidate sample, it indicates that the current repair welding task is under a more severe thin-walled heat dissipation-limited condition than any known sample in the sample library. Directly applying the sample parameters could easily lead to burn-through or severe deformation due to excessive heat input. The local minimum wall thickness data comes from the aforementioned blue light scanning and feature extraction steps, and the local minimum wall thickness of each candidate sample is the known data recorded in the sample library. When any of the above conditions is triggered, it is determined that the cavity to be repaired exceeds the coverage of the welder's repair welding sample library. When the number of candidate samples is zero, automatic repair welding stops and a manual confirmation command is output; when the number of candidate samples is greater than zero, the complete set of repair welding parameters is no longer directly generated based on the conventional sample matching process, but instead, the process switches to the subsequent safety verification mode.

[0150] Step 2: When the number of candidate samples is greater than zero and the cavity to be repaired exceeds the coverage area, control the welding robot to use the verification parameters located in the lower half of the corresponding process evaluation range with effective heat input per unit length, perform verification welds at the bottom layer of the cavity to be repaired, and collect the molten pool state, temperature state and actual weld morphology of the verification welds.

[0151] Once it is confirmed that the current task exceeds the coverage area and the number of candidate samples is greater than zero, a low-risk exploratory welding strategy is implemented. The validation parameters are not optimal process parameters for a specific defect, but rather tentative parameters determined based on conservative principles to test the response under testing conditions. Specifically, the core of configuring the validation parameters is to limit the effective heat input per unit length, placing it in the lower half of the corresponding process qualification range. The process qualification range (PQR) is a parameter safety window determined based on the blade material and welding process qualification tests. The lower half typically refers to a parameter range with lower heat input, shallower penetration, but higher safety. For example, if the allowable heat input range for a material is 0.5 kJ / cm to 1.2 kJ / cm, then the heat input corresponding to the validation parameters could be selected as 0.6 kJ / cm or 0.7 kJ / cm. This low heat input strategy can minimize the risk of excessive heat accumulation or burn-through under unknown conditions. The welding robot performs validation welds at the bottom layer of the repair cavity, i.e., only performing short-distance, small-volume trial welds in the root region where heat dissipation is most difficult, rather than completing the entire cavity filling.

[0152] During the verification weld execution process, multi-source feedback data is simultaneously collected to evaluate the actual welding effect. The molten pool state includes parameters such as molten pool width, molten pool length, molten pool centroid position, and wetting angle, acquired using an industrial camera or high-speed vision sensor mounted at the end of the welding torch. Temperature parameters include real-time temperature distribution in front of, behind, and to the sides of the molten pool, acquired using an infrared thermal imager or a dual-color infrared thermometer. The actual weld morphology includes surface forming characteristics such as weld width, weld height, reinforcement height, and surface waviness after cooling; defects such as porosity, undercut, and lack of fusion are obtained through visual inspection or non-destructive testing. By collecting data on the dynamic behavior of the molten pool, thermal history, and the final weld quality throughout the entire process, objective quantitative evidence is provided for subsequent decision-making.

[0153] Step 3: When the molten pool state and temperature state are both within the target state range determined based on the candidate samples, and the actual weld bead morphology meets the preset requirements for weld bead width, height and continuity, update the subsequent repair welding parameters and repair welding trajectory according to the actual parameters and actual weld bead morphology of the verified weld bead.

[0154] Based on the measured results of the verified weld, a decision is made on whether to return control to the automated planning system. First, the collected molten pool and temperature states are compared with the target state range determined based on candidate samples. The target state range is a statistical interval of molten pool and temperature extracted from existing candidate samples. Although the overall similarity of the candidate samples may be insufficient (triggering coverage determination), the high-quality samples included can still reflect the normal physical process window under this type of material or similar geometry. For example, the target state range may specify that the molten pool width should be between 3 mm and 5 mm, and the temperature behind the molten pool should be below 600 ℃. If the real-time state of the verified weld is within these ranges, it indicates that under the current low thermal input parameters, the thermal process is not out of control, and the material response is within expectations.

[0155] Secondly, the geometric conformity of the actual weld bead morphology is checked. The preset weld bead width and height refer to the weld bead dimensions expected to be achieved based on the size of the repair cavity and the filling strategy. The continuity requirement refers to a smooth weld bead surface without interruptions or significant abrupt changes in width. If the actual weld bead morphology meets these preset requirements, such as a weld bead width deviation within ±0.5 mm and a weld bead continuously covering the predetermined path, it proves that the robot can execute the process parameters well, and that the parameter combination is suitable for the current physical state of the blade.

[0156] Under this premise, the subsequent repair welding parameters and repair welding trajectory are updated based on the actual parameters and morphology of the verified weld. This means that the subsequent repair welding strategy will no longer simply rely on the statistical sample mean, but will be dynamically corrected based on the actual verified physical anchor points. For example, if the verified weld has a good morphology at a current of 80 A and a voltage of 10 V, the subsequent repair welding parameters can be fine-tuned based on this. At the same time, based on the actual weld width and height, the number of filling layers and weld spacing in the remaining area are recalculated to generate a repair welding trajectory that better matches the actual deposition efficiency.

[0157] Step 4: Otherwise, stop automatic welding and output a manual confirmation command.

[0158] If any of the above requirements are not met in the weld pool condition, temperature condition, or actual weld morphology, the current automatic welding system is deemed unable to safely complete the task. In this case, the automatic welding process is immediately stopped to prevent further expansion of the defect or scrapping of the workpiece, and a manual confirmation instruction is issued. The manual confirmation instruction aims to notify the operator to intervene. This instruction can be implemented through pop-up prompts on the workstation interface, audible and visual alarms, or by sending messages to the management terminal. Upon receiving the instruction, the operator can decide, based on the measured data of the verified weld and on-site observation, whether to adjust the process parameters and try again, or switch to a completely manual welding mode. This establishes a safety closed loop of machine verification and manual backup, ensuring that the welding operation remains under control even in extreme conditions not covered by the sample library.

[0159] The above steps, through setting strict pre-judgment conditions, adopting conservative verification strategies, and employing a decision-making mechanism based on actual measurement feedback, construct a highly robust anomaly handling process. When conventional sample matching logic fails in the face of novel weld repair cavity features, this process avoids the system blindly outputting unreliable parameters. Instead, it explores the physical response of unknown working conditions through low-cost verification of the weld bead. Only when the verification results prove that the current physical conditions allow for automated operation will the weld repair task continue; otherwise, losses are stopped in time and the process is handed over to manual handling. This collaborative approach effectively solves the generalization problem of relying solely on data-driven models when facing out-of-distribution samples, significantly improving the engineering practicality and safety of the automated blade weld repair system.

[0160] The above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A blade repair welding method integrating blue light scanning and welder sample learning, characterized in that, include: Blue light scanning was performed on the blade repair welding cavity after defect removal to reconstruct the target surface of the blade, extract the geometric features and thermal boundary features of the repair welding cavity, and establish a local coordinate system for the repair welding cavity; Collect the welding torch trajectory, welding process parameters, molten pool state and temperature state of skilled welders, convert the welding torch trajectory to the local coordinate system of the corresponding repair welding cavity, and align the welding process parameters, molten pool state and temperature state according to the repair welding filling progress. Combine the repair welding quality to build a welder repair welding sample library. Based on the geometric and thermal boundary characteristics of the cavity to be repaired, a welder's repair sample is matched. The physical reference parameters are determined based on the blade material, local wall thickness, and target weld size. Initial repair parameters are generated by using the parameter difference between the matched welder's repair sample and the corresponding physical reference parameters. Based on the repair welding cavity and the target curved surface, a layered and multi-pass repair welding trajectory is generated, the welding robot is controlled to perform repair welding, and the repair welding parameters or weld sequence is adjusted according to the deviation of the real-time molten pool state and temperature state relative to the corresponding state of the repair welding sample of the matched welder. After completing the predetermined weld bead, perform a blue light scan again, and update the subsequent weld trajectory and weld parameters based on the difference between the actual weld surface and the target surface until the weld is completed.

2. The method according to claim 1, characterized in that, The process of reconstructing the target surface of the blade and establishing a local coordinate system for the weld repair cavity includes: The target surface is fitted with the scanning point cloud of the complete blade surface around the welding repair cavity. The centroid of the projection of the welding repair cavity on the target surface is taken as the origin of the coordinate system. The normal of the target surface at the centroid of the projection is taken as the height direction. The main extension direction of the projection contour of the welding repair cavity is taken as the first tangential direction. The local coordinate system of the welding repair cavity is established. The geometric features include the opening area, maximum depth, target filling volume, major-minor axis ratio, and boundary curvature of the welding cavity. The thermal boundary features include the minimum wall thickness, average wall thickness, shortest distance to the free edge of the blade, and shortest distance to the cooling hole in the region where the welding cavity is located.

3. The method according to claim 1, characterized in that, The welding fill progress is determined according to the following formula: ； in, Let t be the progress of the weld filler. The cumulative effective weld volume formed up to time t. The target filling volume for the corresponding weld repair cavity; The cumulative effective deposition volume is determined based on the cross-sectional area of ​​the welding wire, the wire feeding speed, the wire feeding time, and the pre-calibrated effective deposition coefficient, and the welding process parameters, molten pool state, and temperature state in different samples are matched according to the same repair welding filling progress.

4. The method according to claim 1, characterized in that, The matched welder repair samples include: Samples that do not match the material type, welding wire material, or heat treatment status of the blade to be repaired are screened out from the welder's repair sample library; The remaining samples are standardized for target filling volume, maximum depth, aspect ratio, boundary curvature, minimum wall thickness, and shortest distance to cooling holes. The samples are then sorted according to the differences between each standardized feature and the corresponding features of the cavity to be repaired. Samples with qualified welding quality and feature differences not exceeding a preset difference threshold are selected from the ranking results as candidate samples. The sample weights are determined based on the feature differences and welding quality of the candidate samples. When the feature differences increase, the corresponding sample weight decreases. The sample weight of high-quality samples is greater than that of usable samples. The sample weights of each candidate sample are normalized so that the sum of the sample weights of each candidate sample is 1.

5. The method according to claim 4, characterized in that, The physical reference parameters include at least the reference welding current, reference arc voltage, reference welding speed, and reference wire feed speed, and the physical reference parameters satisfy the following effective heat input per unit length: ； Within the process evaluation range corresponding to the blade material and local wall thickness, among which, Effective heat input per unit length For the effective thermal efficiency of the electric arc, Arc voltage For welding current, For welding speed; The initial welding parameters are determined according to the following formula: ； In the formula, These are the initial welding parameters corresponding to the filling progress. The physical reference parameters for the blade to be repaired are as follows: and The first The actual welding process parameters and their corresponding physical reference parameters for each candidate sample For the first The sample weights of each candidate sample.

6. The method according to claim 4, characterized in that, The generation of the layered and multi-pass welding trajectory includes: Based on the depth of the repair welding cavity and the preset single-layer deposition height, the repair welding cavity is divided into a bottom layer, at least one intermediate layer, and a surface layer; First, generate a contour weld bead extending along the boundary of the repair weld cavity at the bottom layer, and then generate an internal filler weld bead within the area enclosed by the contour weld bead; change the extension direction of the internal filler weld bead between adjacent intermediate layers; generate a shaping weld bead on the surface layer that is consistent with the local orientation of the target surface. The spacing between adjacent welds is determined based on the weld width and preset overlap rate in the candidate samples, and the spacing between adjacent welds is reduced at locations where the curvature of the repair cavity increases or the width decreases.

7. The method according to claim 5, characterized in that, Based on the molten pool state, temperature state and sample weight of each candidate sample at the corresponding filling progress, the target state range of the blade to be repaired at the corresponding filling progress is determined. The target state range includes the target ranges corresponding to the molten pool width, molten pool length, molten pool centroid offset, molten pool front temperature, molten pool rear temperature, and temperature difference between the two sides of the molten pool, wherein the first... Item status in fill progress The target value is determined according to the following formula: ； in, Let i be the target value of the i-th state. For the first The corresponding state values ​​of each candidate sample under the same filling progress. For the first The sample weights of each candidate sample; The upper and lower limits of the target state range are determined based on the degree of dispersion of the corresponding state values ​​of the candidate samples relative to the target value.

8. The method according to claim 7, characterized in that, Adjust the welding parameters or weld sequence according to the real-time molten pool condition and temperature condition, including: When the real-time weld pool width is higher than the upper limit of the corresponding target state range, reduce the welding current, increase the welding speed, or reduce the oscillation amplitude. When the real-time weld pool width is lower than the lower limit of the corresponding target state range and no continuous wetting area is formed on the side wall of the repair welding cavity, increase the welding current, reduce the welding speed, or extend the dwell time of the welding torch at the corresponding side wall. When the temperature behind the real-time molten pool is higher than the upper limit of the corresponding target state range, the subsequent welds adjacent to the current weld are stopped, and the next weld is selected from the unwelded welds whose corresponding area temperature is lower than the upper limit of the inter-pass temperature. If there are no unwelded welds whose corresponding area temperature is lower than the upper limit of the inter-pass temperature, the re-welding is suspended until the temperature of at least one unwelded weld area drops below the upper limit of the inter-pass temperature.

9. The method according to claim 7, characterized in that, The cavity to be repaired is determined to be outside the coverage of the welder's repair sample library when at least one of the following conditions is met: the number of candidate samples is less than the preset number of samples; the number of candidate samples is greater than zero and the minimum feature difference among the candidate samples is greater than the preset difference threshold; the number of candidate samples is greater than zero and the local minimum wall thickness of the cavity to be repaired is lower than the minimum value among the local minimum wall thicknesses of each candidate sample. When the number of candidate samples is zero, automatic welding is stopped and a manual confirmation command is output; when the number of candidate samples is greater than zero and the cavity to be welded exceeds the coverage range, the welding robot is controlled to use the effective heat input per unit length to verify the parameters located in the lower half of the corresponding process evaluation range, and to perform verification welds at the bottom layer of the cavity to be welded, and to collect the molten pool state, temperature state and actual weld morphology of the verification welds. When the molten pool state and temperature state are both within the target state range determined based on candidate samples, and the actual weld bead morphology meets the preset weld bead width, height and continuity requirements, the subsequent repair welding parameters and repair welding trajectory are updated according to the actual parameters and actual weld bead morphology of the verified weld bead; otherwise, automatic repair welding is stopped and a manual confirmation command is output.

10. The method according to any one of claims 1-9, characterized in that, The re-scanning with blue light includes performing interlayer scanning on the weld repair area when a fill layer is completed, a preset number of weld beads are completed, or the actual weld bead height is detected to exceed the preset weld bead height range. The actual weld repair surface obtained by interlayer scanning is registered to the target surface. The area where the target surface is higher than the actual weld repair surface and the height difference is greater than the preset underfill threshold is determined as the remaining filling area. The number of subsequent filling layers, the number of weld beads, the weld bead spacing and the wire feed speed are updated according to the area, depth and local curvature of the remaining filling area. The area where the actual weld repair surface is higher than the target curved surface and the height difference is greater than the preset overfill threshold is defined as the overfill area. Subsequent weld beads are stopped in the overfill area, and the overfill area is reserved as the machining area after the weld repair.

Images