Intelligent welding system for hydraulic support structural member

Abstract

The present application relates to the technical field of intelligent welding, and particularly relates to a hydraulic support structural member intelligent welding system, which is based on a hydraulic support structural member welding robot and further comprises: a scanning modeling module, which is used for scanning a current assembly area of a to-be-welded hydraulic support after assembly of each batch of parts of the to-be-welded hydraulic support is completed, and generating a three-dimensional model; an identification comparison module, which is used for locally registering and comparing the three-dimensional model with a pre-stored design drawing model, obtaining a deviation vector field representing a current assembly error, and generating a deviation item list; an adaptive compensation module, which is used for dynamically deforming and pre-compensating weldable weld seams and optimizing a welding sequence, and generating compensated welding parameters; and a welding module, which is used for generating a welding program executable by a welding robot, and enabling the welding robot to perform welding work. The present application solves the technical problems of the prior art, such as dependence on manual operation, low efficiency, high risk and poor quality.

Description

Technical Field

[0001] This invention relates to the field of intelligent welding technology, and in particular to an intelligent welding system for hydraulic support structural components. Background Technology

[0002] Currently, hydraulic support structural robots are mainly divided into two categories in the industry: fixed welding robots without positioning mechanisms and welding robots equipped with positioning mechanisms. These types of equipment generally adopt a structure combining a 6-axis or 7-axis robot body with external axes. The operation mode is mainly manual teaching programming and offline programming, requiring operators to manually set the welding sequence of each weld pass and adjust the working angle of the robot arm.

[0003] In actual welding processes, assembly errors are unavoidable in hydraulic support structural components. Existing welding robots must be manually taught, with the robotic arm inserted into the box-shaped structure point by point to verify coordinate data and correct weld dimensional deviations caused by assembly errors. Simultaneously, manual verification of the robotic arm's spatial movement trajectory is required to prevent collisions during operation. Therefore, manual programming is significantly affected by the operator's skill level, resulting in low efficiency and difficulty in ensuring consistent welding quality. The high risk of robotic arm collisions during manual teaching further reduces production efficiency. Furthermore, due to assembly errors in structural components, the design blueprints offer limited guidance for programming, necessitating manual verification to correct deviations, thus increasing additional operating costs.

[0004] Therefore, there is an urgent need for an intelligent welding system for hydraulic support structural components. Summary of the Invention

[0005] (a) Technical problems to be solved

[0006] In view of the above-mentioned shortcomings of the prior art, the present invention provides an intelligent welding system for hydraulic support structural components, which solves the technical problems of low welding quality and low production efficiency in the prior art.

[0007] (II) Technical Solution

[0008] To achieve the above objectives, the main technical solutions adopted by the present invention include:

[0009] This application proposes an intelligent welding system for hydraulic support structural components, which, based on a welding robot for hydraulic support structural components, further includes:

[0010] The scanning modeling module is used to scan the current assembly area of ​​the hydraulic support to be welded after each batch of parts is assembled, and generate a 3D model.

[0011] The identification and comparison module is used to perform local registration and comparison between the three-dimensional model and the pre-stored design drawing model, obtain the deviation vector field characterizing the current assembly error, and generate a list of deviation items; the list of deviation items includes weldable welds with welding space under the current assembly area and welds that are temporarily unweldable due to adjacent parts not being assembled or path obstruction.

[0012] An adaptive compensation module is used to perform dynamic deformation pre-compensation and weld sequence optimization on the weldable weld seam based on the deviation vector field and the deviation item list, and generate compensated welding parameters; the welding parameters include the compensated three-dimensional welding trajectory inside each weldable weld seam and the optimized weld sequence between weldable weld seams;

[0013] The welding module is used to generate a welding program that can be executed by the welding robot based on the compensated welding parameters, so that the welding robot can perform welding operations.

[0014] Optionally, in some embodiments of this application, the identification and comparison module specifically includes:

[0015] The model acquisition unit is used to acquire the theoretical model of the current assembly area corresponding to the three-dimensional model from the pre-stored design drawings based on the three-dimensional model.

[0016] The deviation calculation unit is used to perform rigid body registration between the three-dimensional model and the theoretical model using the iterative nearest point algorithm, establish a point-to-point correspondence between the three-dimensional model and the theoretical model, and obtain the coordinate difference and normal vector angle of each pair of corresponding points; based on the coordinate difference and normal vector angle of each pair of corresponding points, the three-dimensional position deviation and normal vector angle deviation of each pair of corresponding points are obtained.

[0017] The field generation unit is used to take the three-dimensional position deviation and normal vector angle deviation of each point as known sample points for radial basis function interpolation, obtain the continuous deviation covering the current assembly area through radial basis function interpolation, and generate the deviation vector field.

[0018] Optionally, in some embodiments of this application, the identification and comparison module further includes:

[0019] The assembly dependency judgment unit is used to mark the welds in the current assembly area that cannot be welded because the adjacent parts have not yet been assembled as assembly-unwelded welds, according to the pre-set batch welding sequence plan.

[0020] The spatial accessibility determination unit is used to determine whether the weld has welding space based on the three-dimensional model and the deviation vector field; if there is no welding space, it is marked as a weld that cannot be welded temporarily.

[0021] The list generation unit is used to combine the assembly-temporarily unweldable welds and the space-temporarily unweldable welds into a list of temporarily unweldable welds, and to include the remaining welds that have welding space and no assembly dependency into a list of weldable welds.

[0022] Optionally, in some embodiments of this application, the spatial reachability determination unit includes:

[0023] The information extraction subunit is used to obtain weld information for each weld seam based on the three-dimensional model and the deviation vector field; the weld seam information includes the weld seam's position coordinates, normal offset, and groove geometric parameters;

[0024] The interference volume ratio acquisition subunit is used to obtain the interference volume ratio when the welding torch approaches each weld seam in the optimal posture using a ray projection algorithm. The interference volume ratio is the percentage of the volume of the welding torch motion envelope intersecting with the surrounding obstacles to the total volume of the welding torch motion envelope.

[0025] The confidence assessment subunit is used to input the interference volume ratio and weld information of each weld into a pre-trained accessibility prediction neural network to obtain the accessibility confidence of each weld. When the accessibility confidence is greater than a preset threshold, it is determined that there is space for welding.

[0026] Optionally, in some embodiments of this application, the optimal posture in the interference volume ratio acquisition subunit is:

[0027] Within the preset ranges of the welding torch working angle and travel angle, the welding torch posture is the one that minimizes the ratio of the interference volume between the welding torch motion envelope and surrounding obstacles.

[0028] Optionally, in some embodiments of this application, the adaptive compensation module includes:

[0029] The trajectory compensation unit is used to obtain the dynamic deformation amount of the weldable welds in the deviation item list according to the deviation vector field using the welding deformation prediction model, and to apply the dynamic deformation amount in reverse on the theoretical welding trajectory to generate the compensated three-dimensional welding trajectory inside each weldable weld.

[0030] The sequence optimization unit is used to solve the globally optimal weld sequence that minimizes the overall residual deformation by using a graph optimization algorithm, based on the process constraints of batch welding of hydraulic supports and with the heat-affected coupling strength between welds as the weight, according to the process constraints of batch welding of hydraulic supports, and to generate the optimized weld sequence.

[0031] Optionally, in some embodiments of this application, the welding deformation prediction model in the trajectory compensation unit is a neural network model trained based on historical welding measurement data, and the neural network model includes:

[0032] The input layer is used to receive the local gap of each weldable weld and the preset heat input amount; the local gap is obtained based on the three-dimensional model;

[0033] The hidden layer consists of at least three fully connected layers and employs ReLU activation and Dropout regularization.

[0034] The output layer is used to output the dynamic deformation evolution curve of the weldable weld during and after welding. The curve is plotted with time on the horizontal axis and the distribution of dynamic deformation along the weld length on the vertical axis.

[0035] Optionally, in some embodiments of this application, the sequence optimization unit includes:

[0036] The coupling strength acquisition subunit is used to acquire the spatial distance, preset heat input and preset heat dissipation conditions between every two welds in the weldable weld according to the three-dimensional model, and to acquire the heat-affected coupling strength between every two welds according to the spatial distance, preset heat input and preset heat dissipation conditions.

[0037] The graph construction sub-unit is used to construct a weighted weld sequence relationship graph with each weldable weld as a node and the heat-affected coupling strength as the edge weight.

[0038] The sequential solution sub-unit is used to employ a genetic algorithm to minimize the overall residual deformation during and after batch welding. Under the premise of satisfying the process constraints of batch welding, the globally optimal traversal order of the weighted weld bead sequence relationship graph is obtained as the optimized weld bead sequence.

[0039] Optionally, in some embodiments of this application, the system further includes an obstacle avoidance detection module;

[0040] The obstacle avoidance detection module is specifically used to collect the spatial distance between the robotic arm of the welding robot and the reserved positions of assembled and unassembled parts in real time during the welding operation. When the spatial distance is less than the safety threshold, the robot automatically adjusts its movement trajectory or suspends the operation.

[0041] Optionally, in some embodiments of this application, the system further includes a model update module, the model update module comprising:

[0042] The storage unit is used to store the 3D model, deviation vector field, list of weldable welds and compensated welding parameters of each batch of welding operations into the database.

[0043] The training unit is used to periodically retrain the reachability prediction neural network and the welding deformation prediction model using historical data in the database, so that the welding accuracy of the system for subsequent batches of hydraulic supports of the same model can be gradually improved.

[0044] (III) Beneficial Effects

[0045] The beneficial effects of this invention are as follows: The intelligent welding system for hydraulic support structural components of this invention employs a scanning modeling module to perform three-dimensional reconstruction of each batch of assembled hydraulic supports to be welded; an identification and comparison module to locally register the three-dimensional model with the design drawing model and generate a deviation vector field and a list of deviation items; an adaptive compensation module to dynamically deform and pre-compensate weldable welds and optimize weld sequence based on deviation information; and a welding module to automatically generate compensated welding programs. Compared with existing technologies, this system can quantify assembly errors in real time and distinguish between weldable and temporarily non-weldable welds, providing targeted compensation for welding trajectories and sequences. This effectively avoids welding defects and collision risks caused by component assembly deviations or path obstructions, while significantly reducing the workload of manual teaching and rework adjustments. It achieves the beneficial effects of improving welding quality consistency, reducing scrap rates, enhancing the system's adaptability to complex assembly conditions, and significantly improving the automation and intelligence level of hydraulic support structural component welding production. Attached Figure Description

[0046] Figure 1 This is a schematic diagram of the intelligent welding system for hydraulic support structural components according to an embodiment of this application;

[0047] Figure 2 This is a schematic diagram of the structure of a spatial accessibility determination unit according to an embodiment of this application;

[0048] Figure 3 This is a schematic diagram of the structure of a sequence optimization unit according to an embodiment of this application;

[0049] Figure 4 This is a schematic diagram of the structure of a welding robot according to an embodiment of this application. Detailed Implementation

[0050] To better explain and facilitate understanding of the present invention, the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0051] In existing technologies, welding robots for hydraulic support structures are mainly divided into two categories: fixed and those with positioners. They typically employ a 6-axis or 7-axis robot body plus external axes, and their operation primarily relies on manual teaching programming and offline programming. Operators need to manually set the welding sequence for each weld pass and adjust the robotic arm's working angle. However, due to unavoidable assembly errors during the actual welding process of hydraulic support structures, existing welding robots must rely on manual teaching. The robotic arm is inserted point-by-point into the box-shaped structure to verify coordinate data and correct weld dimensional deviations, while simultaneously verifying the spatial movement trajectory to avoid collisions. This manual programming method is significantly affected by the operator's skill level, resulting in low work efficiency and difficulty in ensuring consistent welding quality. The high risk of robotic arm collisions during manual teaching further reduces production efficiency. Furthermore, due to assembly errors, the design blueprints have limited guiding role in programming, necessitating repeated manual verification to correct deviations, thus increasing additional operating costs.

[0052] To address the aforementioned issues, this application proposes an intelligent welding system for hydraulic support structural components. Based on a welding robot, it integrates a scanning and modeling module, a recognition and comparison module, an adaptive compensation module, and a welding module. First, the scanning and modeling module generates a 3D model of the area to be welded after each batch of assembly. Then, the recognition and comparison module locally registers the 3D model with the design drawing model, obtaining a deviation vector field characterizing assembly errors and generating a list of deviation items including weldable and temporarily unweldable welds. Next, the adaptive compensation module, based on the deviation vector field and the list, performs dynamic deformation pre-compensation and weld sequence optimization for weldable welds, outputting the compensated 3D welding trajectory and the globally optimal weld sequence. Finally, the welding module automatically generates a welding program executable by the robot. This system can quantify assembly errors in real time, intelligently select weldable welds, and dynamically adaptively compensate for the welding trajectory and sequence, significantly reducing welding defects and collision risks, and greatly improving the automation and intelligence level of hydraulic support structural component welding.

[0053] It is important to understand that hydraulic support structural components typically employ a batch welding process. This means that, according to the design drawings and assembly sequence, numerous components (such as large structural components like top beams, shield beams, connecting rods, and bases, as well as various connecting lugs, stiffeners, and mounting plates) are divided into multiple batches for sequential assembly and welding. The first batch of parts is first spot-welded or partially welded to form a preliminary skeleton or local assembly. Then, the next batch of parts is assembled on this basis, followed by corresponding supplementary welding or full welding. This process is repeated layer by layer until all the welding of the entire structural component is completed.

[0054] To better understand the above technical solutions, exemplary embodiments of the present invention will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present invention are shown in the drawings, it should be understood that the present invention can be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that the present invention can be understood more clearly and thoroughly, and that the scope of the present invention can be fully conveyed to those skilled in the art.

[0055] Figure 1 This is a structural schematic diagram of an intelligent welding system for hydraulic support components according to an embodiment of this application. Figure 1 As shown, this intelligent welding system is used in welding robots for hydraulic support structures (such as...). Figure 4 Based on (as shown), it also includes:

[0056] The scanning modeling module is used to scan the current assembly area of ​​the hydraulic support to be welded after each batch of parts is assembled, and generate a 3D model.

[0057] Specifically, the scan modeling module includes:

[0058] The 3D scanning unit is used to automatically scan the current assembly area from multiple angles and at multiple stations after each batch of parts of the hydraulic support to be welded is assembled, using a structured light scanner or lidar, to obtain the original point cloud data covering the area to be welded and the surfaces of adjacent parts; the scanning process is executed according to the preset scanning path and overlap rate to ensure that there are sufficient registration features between the point clouds of adjacent stations.

[0059] The point cloud preprocessing unit, connected to the 3D scanning unit, is used to filter and denoise the raw point cloud data, remove outliers, and simplify the data to generate a clean and uniform preprocessed point cloud. The filtering and denoising includes statistical filtering or radius filtering, and the data simplification adopts the voxel grid downsampling method.

[0060] The point cloud registration unit is used to globally register the preprocessed point clouds acquired from multiple stations using an iterative nearest point algorithm, eliminating coordinate system differences between different scanning perspectives and fusing them to form an overall point cloud model covering the complete surface of the current assembly area.

[0061] The surface reconstruction unit, connected to the point cloud registration unit, is used to perform triangulation processing on the overall point cloud model using Poisson surface reconstruction or rolling ball algorithm to generate an editable three-dimensional mesh model containing geometric shape, surface normal and local curvature information, which is then output to the recognition and comparison module as the three-dimensional model.

[0062] The model coordinate system alignment unit is used to initially align the three-dimensional mesh model with the global coordinate system of the pre-stored design drawing model through principal component analysis or rigid body transformation based on reference feature points (such as process holes, positioning pin holes or pre-welded reference blocks), so that the subsequent identification and comparison module can obtain a unified registration reference.

[0063] For example, taking the first batch of assembly of the hydraulic support top beam as an example, this batch has completed the spot welding and fixing of the main web plate, upper cover plate, lower cover plate, and internal transverse stiffeners, forming a box-shaped frame. After the scanning and modeling module is started, the 3D scanning unit first controls the robotic arm equipped with a structured light scanner to perform multi-angle scanning of the top beam according to the preset path: first scanning the external upper surface, lower surface, and both outer walls, then the robotic arm carrying the scanner extends into each compartment inside the box-shaped frame to perform a detailed scan of the weld area where the stiffeners and web plate meet. Each station scans for about 2-3 seconds, and the raw point cloud data is collected in real time. Next, the point cloud preprocessing unit uses statistical filtering to remove outliers caused by dust reflection, and then uses voxel downsampling to homogenize the point cloud density, reducing the amount of data while retaining key geometric features. Subsequently, the point cloud registration unit uses the iterative nearest point algorithm to register and fuse the point clouds of multiple stations one by one, eliminating the coordinate differences between different scanning perspectives, and forming a complete point cloud model covering the entire current assembly area. The surface reconstruction unit then uses the Poisson surface reconstruction algorithm to triangulate the point cloud, generating an editable 3D mesh model. This model clearly reflects the actual assembly state—for example, a gap of 2.5mm between a stiffening plate and the cover plate due to spot welding positioning error, while the design drawing specifies 2.0mm; this deviation is accurately captured. Finally, the model coordinate system alignment unit identifies the process holes at both ends of the top beam as reference feature points and performs rigid body transformation to align the 3D mesh model with the pre-stored design drawing model, providing a unified registration benchmark for subsequent identification and comparison modules. The entire process is fully automated, taking approximately 3 minutes, and outputting a 3D model with an accuracy of ±0.3mm, enabling subsequent modules to perform precise compensation based on actual assembly errors.

[0064] The identification and comparison module is used to perform local registration and comparison between the 3D model and the pre-stored design drawing model, obtain the deviation vector field that represents the current assembly error, and generate a list of deviation items. The list of deviation items includes weldable welds with welding space in the current assembly area and welds that are temporarily unweldable due to adjacent parts not being assembled or path obstruction.

[0065] Specifically, the identification and comparison module includes:

[0066] The model acquisition unit is used to obtain the theoretical model of the current assembly area corresponding to the 3D model from the pre-stored design drawings based on the 3D model.

[0067] In the specific implementation process, the model acquisition unit automatically retrieves the theoretical CAD model of the current assembly area from the system's pre-stored design drawing library based on the part number and assembly batch of the 3D model. For example, if the header information of the 3D model records "Top Beam - First Batch - Stiffener Welding Stage", the model acquisition unit will match the design drawing model of the same version and extract the area consistent with the spatial range of the measured model (such as the 2nd to 4th compartments in the middle of the top beam), outputting it as a theoretical model. This theoretical model includes the ideal geometric dimensions of each part, the theoretical weld position, the bevel angle, and the nominal relative positional relationship between parts.

[0068] The deviation calculation unit is used to perform rigid body registration between the 3D model and the theoretical model using the iterative nearest point algorithm, establish a point-to-point correspondence between the 3D model and the theoretical model, and obtain the coordinate difference and normal vector angle of each pair of corresponding points; based on the coordinate difference and normal vector angle of each pair of corresponding points, the 3D position deviation and normal vector angle deviation of each pair of corresponding points are obtained.

[0069] This deviation calculation unit employs an iterative nearest-point algorithm to perform fine rigid body registration between the measured model and the theoretical model, eliminating minor pose deviations remaining after initial alignment. For example, using the theoretical model as a reference, iteratively searches for the closest point pair in the measured model to the theoretical model, optimizing the rigid body transformation matrix (rotation + translation) by minimizing the sum of squared distances between point pairs. After approximately 50 iterations, convergence occurs, with the registration error less than 0.1 mm. At this point, the measured model and the theoretical model achieve optimal spatial overlap.

[0070] After registration, for each point on the theoretical model (e.g., the dense set of points in the weld area at the junction of the stiffener and the cover plate), its corresponding point is found in the measured model (using nearest neighbor search). The coordinate difference and the angle between the normal vectors of each pair of corresponding points are calculated. For example, at a point in the upper right corner of the stiffener, the theoretical coordinates are (100.0, 200.0, 50.0) mm, and the normal vector is (0, 0, 1); the measured corresponding point has coordinates of (101.2, 200.0, 50.0) mm, and the normal vector is (0.02, 0, 0.9998). This results in a three-dimensional positional deviation of (1.2, 0, 0) mm and a normal vector angle of approximately 1.15°. This point reflects a 1.2 mm rightward offset of the stiffener and a slight surface tilt.

[0071] Therefore, the three-dimensional position deviations and normal vector angle deviations of all corresponding points constitute a set of discrete sample data, and each sample point contains coordinates and the corresponding deviation vector.

[0072] The field generation unit is used to take the three-dimensional position deviation and normal vector angle deviation of each point as known sample points for radial basis function interpolation, obtain the continuous deviation covering the current assembly area through radial basis function interpolation, and generate the deviation vector field.

[0073] In other words, the core task of the field generation unit is to transform a series of discrete sample points (each containing spatial coordinates and its corresponding 3D position deviation and normal vector angle deviation) obtained by the deviation calculation unit into a continuous vector field covering the entire current assembly area. Specifically, the 3D position deviation is decomposed into three independent directional components, and interpolated along with the normal vector angle deviation. For one directional component, the field generation unit constructs a smooth interpolation function using the spatial positions of all discrete sample points as known points and the deviation value at that point as a known value. The basic principle of this function is that the deviation value at any point in space is equal to the weighted sum of the influence of all known sample points on that point. The magnitude of the influence of each sample point depends on the distance between the point to be calculated and the sample points—the closer the distance, the greater the influence; the farther the distance, the smaller the influence, and the influence decays smoothly with distance (this rule is described by the radial basis function). Simultaneously, to ensure that areas far from the sample points do not exhibit abnormal fluctuations, a smooth background trend term is superimposed. By solving a system of linear equations, the weighting coefficients and background trend term coefficients for each sample point can be determined, ensuring that the interpolation function is precisely equal to the measured deviation value at all known sample points and smoothly transitions between points. After solving, a continuous spatial deviation field is obtained: by inputting any theoretical coordinate within the assembly area, the actual 3D position deviation and normal vector angle deviation of that point can be quickly queried. Finally, the field generation unit integrates the position deviation field and angle deviation field in the three directions into a complete deviation vector field, stored in voxel mesh or coefficient form, for real-time use by the subsequent adaptive compensation module.

[0074] For example, suppose that within a compartment of the top beam, the scan only obtains deviation information at three discrete points: point A, where the stiffener is offset to the right by 1.2 mm; point B, by 1.1 mm; and point C, by 1.3 mm. The field generation unit, based on the spatial positions and offset values ​​of these three points, calculates the continuous offset at any position within the entire compartment through radial basis function interpolation. For instance, at a position exactly between points A and B, the system will comprehensively consider the influence of both points A and B (the closer the distance, the greater the weight), automatically calculating the offset at that point to be approximately 1.15 mm. In this way, what were originally just three isolated deviation data points become a smooth and continuous deviation field. Subsequently, when the welding robot plans the weld trajectory, regardless of where the path point falls, the actual offset value of that point can be queried in real time, thus enabling precise compensation.

[0075] Therefore, the identification and comparison module of this application embodiment can automatically and precisely register the scanned 3D model with the design drawing model. By calculating the coordinate difference and the angle between the normal vectors point by point, a continuous deviation vector field covering the entire assembly area is generated, thereby accurately quantifying the assembly error generated after each batch of welding.

[0076] Furthermore, the identification and comparison module also includes:

[0077] The assembly dependency judgment unit is used to mark the welds in the current assembly area that cannot be welded because the adjacent parts have not yet been assembled as temporarily unweldable welds, according to the pre-set batch welding sequence plan, so as to avoid the welding robot trying to weld joints that do not yet exist.

[0078] In detail, in the batch welding process of hydraulic support structural components, the entire welding process is divided into multiple batches, with each batch assembling and welding only a portion of the parts. For example, the welding sequence plan for a certain type of top beam stipulates: the first batch assembles the main web plate, upper and lower cover plates, and transverse stiffeners, and welds the connecting welds between these parts; the second batch assembles the longitudinal stiffeners and each ear plate, at which point the welds between the longitudinal stiffeners and the web plate, as well as the fillet welds between the ear plates and the cover plates, can be welded; the third batch assembles the mounting plates and lifting rings, and then welds the remaining welds. Therefore, when scanning and identifying after the first batch is completed, although the welds within the first batch (such as the vertical welds between the stiffeners and the cover plates) are ready for welding, the welds involving the parts in the second and third batches (such as the ear plate welds) cannot be welded at all because the ear plates have not yet been installed.

[0079] Specifically, the workflow of the assembly dependency judgment unit is as follows:

[0080] First, the system stores a batch welding sequence plan file for the current hydraulic support model. This file uses a structured data format (such as XML or a database table) and clearly records the list of parts to be assembled in each batch, the weld identifier on each part, and the batch number of the part that each weld depends on. For example, the record for a fillet weld between an ear plate and a cover plate will indicate that the dependent part is the left ear plate, and this part belongs to the second batch of assembly; this weld can only be made after the second batch of assembly is completed.

[0081] Secondly, after each batch of assembly is completed (e.g., the first batch is spot-welded and fixed), the scanning and modeling module generates a 3D model, and the assembly dependency judgment unit automatically reads the current batch number (e.g., batch 1). Then, it iterates through all welds defined in the design drawing model (each weld has a unique ID). For each weld, it queries the parts it depends on and the assembly batch to which those parts belong. If the batch number of the part is greater than the current batch number, it means that the part has not yet been assembled, and the weld is marked as temporarily unavailable for welding; if the batch number of the part is less than or equal to the current batch number, it means that the part has been assembled, and the weld proceeds to the subsequent spatial reachability judgment.

[0082] For example, after the first batch of top beam assembly is completed, the system checks the vertical weld between the stiffening plate and the cover plate. The parts it depends on are the transverse stiffening plate and the upper cover plate, both of which belong to the first batch, so it is determined that it can proceed to the next stage of evaluation. However, the fillet weld between the left ear plate and the cover plate depends on the left ear plate, which belongs to the second batch and has not yet been assembled. Therefore, it is directly marked as a weld that cannot be welded at this stage and recorded in the deviation item list, prompting the welding robot not to process this weld in this batch. Similarly, the lap weld between the side plate and the web plate depends on the side plate, which belongs to the third batch and is also marked.

[0083] Furthermore, the assembly dependency judgment unit also handles local dependencies in the assembly sequence: some welds connect two parts that belong to the current batch, but one of the parts is assembled later in a sub-step of the current batch (e.g., spot welding the positioning stiffener first, then assembling the backing plate within the same batch). In this case, more granular sub-step information is needed. The system supports defining sub-stages within each batch in the batch welding sequence plan. The assembly dependency judgment unit will also check whether the parts that the weld depends on have been assembled in the current sub-stage. If not, they are marked as temporarily unweldable.

[0084] Finally, the assembly dependency judgment unit outputs a list containing all temporarily unweldable weld identifiers and their reasons (e.g., dependent parts not assembled: left ear plate (second batch)), which is then passed to the list generation unit to form a complete list of temporarily unweldable welds together with the spatially unweldable welds. This unit effectively avoids the robot making empty welds or colliding due to missing parts, and also automatically retains welding tasks for subsequent batches, realizing intelligent management of dynamically adjusting welding content by batch.

[0085] Furthermore, after identifying the welds that are temporarily not weldable in the assembly process, the following modules are also used to identify the welds that are temporarily not weldable in the spatial process:

[0086] The spatial accessibility judgment unit is used to determine whether the weld has space for welding based on the three-dimensional model and the deviation vector field; if there is no space for welding, it is marked as a weld that cannot be welded temporarily.

[0087] like Figure 2 As shown, the spatial accessibility determination unit includes:

[0088] The information extraction subunit is used to obtain the weld information of each weld seam based on the three-dimensional model and the deviation vector field; the weld information includes the weld seam's position coordinates, normal offset, and groove geometric parameters;

[0089] The interference volume ratio acquisition sub-unit is used to obtain the interference volume ratio when the welding torch approaches each weld seam in the optimal posture using the ray projection algorithm. The interference volume ratio is the percentage of the volume of the welding torch motion envelope intersecting with the surrounding obstacles to the total volume of the welding torch motion envelope.

[0090] Specifically, the system needs to pre-create a precise 3D model of the welding robot's end effector (welding torch), including components such as the torch body, nozzle, contact tip, and wire feed tube. Therefore, for a given welding torch posture (defined by the working angle, travel angle, and rotation angle), the spatial area occupied by the welding torch can be calculated, called the "welding torch motion envelope." In actual simulations, the welding torch is usually simplified to a combination of spheres or cylinders to improve computational efficiency.

[0091] For each path point on each weld seam, the system performs a discretized search within preset ranges for the welding torch's working angle and travel angle. The working angle typically refers to the angle between the welding torch axis and the normal to the weld seam surface, ranging from 30° to 60°; the travel angle refers to the tilt angle of the welding torch in the weld seam's forward direction, ranging from 0° to 30°. Under each discrete posture combination, the welding torch model is placed at the path point, aligning the welding wire tip with the actual weld seam position at that point. Then, the interference volume between the welding torch's motion envelope and surrounding obstacles (the surfaces of assembled parts extracted from the 3D model) is calculated.

[0092] Interference volume ratio = (volume of the intersection of the welding torch motion envelope and the obstacle) / (total volume of the welding torch motion envelope) × 100%. The smaller this ratio, the less interference there is and the better the orientation.

[0093] The system iterates through all discrete posture combinations and selects the posture that minimizes the interference volume ratio as the optimal posture for that path point. If the interference volume ratio of multiple postures is 0 (no interference at all), the posture with the working angle closer to the ideal value (e.g., 45°) is further selected as the optimal posture. In other words, within the preset range of the welding torch working angle and the preset range of the travel angle, the welding torch posture that minimizes the interference volume ratio between the welding torch motion envelope and surrounding obstacles is the optimal posture.

[0094] For example, suppose there is a stiffening plate weld located deep inside a box-shaped structure. During the search process, when the welding torch working angle is 50° and the travel angle is 15°, the tail of the welding torch just makes a slight collision with the opposite web plate, with an interference volume ratio of 5%. When the working angle is adjusted to 45° and the travel angle is 10°, the interference volume ratio drops to 0%. This posture is then selected as the optimal one, with an average interference volume ratio of 0.

[0095] The confidence assessment subunit is used to input the interference volume ratio and weld information of each weld into a pre-trained accessibility prediction neural network to obtain the accessibility confidence of each weld. When the accessibility confidence is greater than a preset threshold, it is determined that there is space for welding.

[0096] The reachability prediction neural network employs a multilayer perceptron structure (e.g., 3 hidden layers, 64 neurons per layer), outputting reachability confidence scores between 0 and 1. Training data is derived from measured data recorded in historical welding operations: for each weld seam previously attempted, its aforementioned feature parameters and the actual welding result (successful welding, mid-way collision alarm, welding torch inability to reach, etc.) are recorded. Successfully welded samples are marked with a confidence score of 1, and failed samples are marked with 0. Through extensive training with numerous samples, the network learns a nonlinear mapping from geometric and interference features to actual reachability.

[0097] Furthermore, assuming the preset threshold for reachability confidence is 0.8, when the confidence level output by the neural network is greater than or equal to 0.8, the weld is determined to have welding space; otherwise, it is determined that the space is temporarily unsuitable for welding. The threshold can be dynamically adjusted according to the actual production line requirements (e.g., for high quality requirements, the threshold can be increased to 0.9 to further reduce the risk of collision).

[0098] In this embodiment, the aforementioned spatial accessibility judgment unit calculates the interference volume ratio when the welding torch approaches each weld seam in the optimal posture using a precise ray projection algorithm. It then combines this with a confidence assessment output from an accessibility prediction neural network trained on historical data. This intelligently identifies weld seams that are temporarily unsuitable for welding due to part obstruction, narrow space, or complex geometric constraints. Compared to traditional methods relying on manual experience or simple teaching verification, this unit comprehensively quantifies the dynamic interference risk between the welding torch and surrounding obstacles, effectively avoiding collisions, equipment damage, and welding defects caused by the welding robot's blind attempts. Simultaneously, it pre-screens high-risk weld seams for manual intervention or re-evaluation in subsequent batches, significantly improving the safety, reliability, and overall operational efficiency of the welding process.

[0099] Furthermore, the identification and comparison module also includes:

[0100] The list generation unit is used to combine the temporarily non-weldable welds in the assembly with the temporarily non-weldable welds in the space into a list of temporarily non-weldable welds, and to include the remaining welds that have welding space and no assembly dependence into the list of weldable welds.

[0101] The intelligent welding system in this embodiment also includes: an adaptive compensation module, which is used to perform dynamic deformation pre-compensation and weld sequence optimization processing on weldable welds based on the deviation vector field and the deviation item list, and generate compensated welding parameters; the welding parameters include the compensated three-dimensional welding trajectory inside each weldable weld and the optimized weld sequence between weldable welds;

[0102] The adaptive compensation module includes a trajectory compensation unit and a sequence optimization unit, specifically:

[0103] The trajectory compensation unit is used to obtain the dynamic deformation amount of the weldable welds in the deviation item list according to the deviation vector field using the welding deformation prediction model, and to apply the dynamic deformation amount in reverse on the theoretical welding trajectory to generate the compensated three-dimensional welding trajectory inside each weldable weld.

[0104] The welding deformation prediction model in the trajectory compensation unit is a neural network model trained based on historical welding measurement data. The neural network model includes:

[0105] The input layer receives the local gap and preset heat input for each weldable weld. The local gap, obtained from the 3D model, is the actual distance between the surfaces of the parts to be joined on both sides of the weld. Based on the 3D mesh model generated by the scanning modeling module, the system samples a series of path points at a certain step size (e.g., every 2mm) for the theoretical trajectory of each weldable weld. At each path point, rays are projected along the normal vector of that point to the surfaces of the parts on both sides, and the shortest distance between the two surfaces is calculated; this distance is the local gap at that point. The preset heat input is determined by welding industry parameters, and its calculation formula is: Heat Input = (Welding Voltage × Welding Current) / Welding Speed. The system has a pre-built database of welding process specifications for different weld types (such as fillet welds, butt welds, and bevel welds). Once the identification and comparison module determines the weld information of a weldable weld, the system automatically matches the recommended standard process parameters from the process database and then calculates the preset heat input.

[0106] The process involves obtaining weld information for a weldable joint, determining the basic type based on the angle between the normal vectors of the two surfaces to be welded—if the angle is close to 90°, it's a fillet weld; if it's close to 180°, it's a butt weld; and if one surface overlaps the other, it's a lap weld. This is further refined by combining groove geometry parameters (such as groove angle and root gap). For example, in a butt weld, a groove angle of 30°~45° indicates a V-groove, and 8°~12° indicates a narrow-gap groove; in a fillet weld, if there's a single-sided bevel, it's a single-sided V-groove fillet weld. Based on this, the weld type is determined, and recommended standard process parameters are matched from the process database.

[0107] The hidden layer consists of at least three fully connected layers and employs ReLU activation and Dropout regularization.

[0108] The output layer is used to output the dynamic deformation evolution curves of the weldable weld during and after welding. The curves are plotted with time on the horizontal axis and the distribution of dynamic deformation along the weld length on the vertical axis.

[0109] The trajectory compensation unit in this embodiment of the application uses a neural network model trained based on historical welding measurement data. Taking the actual local gap of the weld and a preset heat input as input, it accurately predicts the dynamic deformation evolution curve along the weld length during and after welding. It then applies this deformation in reverse along the theoretical welding trajectory to generate a compensated three-dimensional welding trajectory. Compared to traditional methods that rely on manual experience for correction or fixed trajectory welding, this unit effectively offsets non-uniform deformation caused by assembly errors, uneven local stiffness, and heat input. This significantly reduces post-weld straightening workload, lowers the incidence of welding defects (such as lack of fusion, undercut, and porosity), and greatly improves the dimensional accuracy and consistency of hydraulic support structural component welding, achieving intelligent dynamic compensation of the welding trajectory.

[0110] The sequence optimization unit is used to solve the globally optimal weld sequence that minimizes the overall residual deformation by using a graph optimization algorithm, based on the process constraints of batch welding of hydraulic supports and with the heat-affected coupling strength between welds as the weight, and to generate the optimized weld sequence.

[0111] like Figure 3 As shown, the sequential optimization unit includes:

[0112] The coupling strength acquisition subunit is used to obtain the spatial distance, preset heat input and preset heat dissipation conditions between each two welds in the weldable weld seam based on the three-dimensional model, and to obtain the heat-affect coupling strength between each two welds based on the spatial distance, preset heat input and preset heat dissipation conditions.

[0113] Specifically, the formula for the thermally affected coupling strength is:

[0114] ;

[0115] Where C is the thermally affected coupling strength, and Q A and Q B These are the preset heat inputs for the two welds, S is the heat dissipation coefficient (i.e., the coefficient for quantifying heat dissipation conditions, with a value range of 0-1, where 0 indicates extremely fast heat dissipation and 1 indicates adiabatic conditions), and d is the spatial distance between the two welds.

[0116] The graph construction sub-unit is used to construct a weighted weld sequence relationship graph with each weldable weld as a node and the heat-affected coupling strength as the edge weight.

[0117] For example, suppose there are three weldable seams A, B, and C. The calculated coupling strength is 0.45 between A and B, 0.36 between B and C, and 0.12 between A and C. The graph construction sub-element treats A, B, and C as three nodes, adding an edge between each pair of welds. The weight of the edge represents the corresponding coupling strength, thus forming a weighted complete graph. This graph visually represents the strength of the heat effect between welds: A and B have the strongest coupling and should avoid continuous welding; A and C have the weakest coupling and their order can be similar. Subsequent solution sub-elements use this graph to search for the optimal weld sequence.

[0118] The sequential solution sub-unit is used to employ a genetic algorithm to minimize the overall residual deformation during and after batch welding. Under the premise of satisfying the process constraints of batch welding, the globally optimal traversal order of the weighted weld bead sequence relationship graph is obtained as the optimized weld bead sequence.

[0119] In the specific implementation process, assume that there are 5 weldable seams in the current batch, numbered W1, W2, W3, W4, and W5. The heat-affected coupling strength between each pair of weld seams has been calculated in the drawing construction sub-units. For example, the coupling between W1 and W2 is very strong (0.45), and the coupling between W4 and W5 is also very strong (0.50), while the coupling of other combinations is weaker. In addition, there are two constraints for the batch welding process: W1 and W2 must be welded before W3 (i.e., W3 cannot be welded in the first two), and W4 and W5 cannot be welded adjacent to each other.

[0120] The sequential solution sub-unit employs a genetic algorithm to search for the welding sequence that minimizes the overall residual deformation. First, it randomly generates a batch of welding sequences satisfying the above constraints as an initial population, such as [W1,W2,W3,W4,W5], [W1,W4,W2,W5,W3], [W2,W1,W4,W3,W5], etc. For each sequence, the algorithm calculates a "deformation cost"—the sum of the coupling strengths of all adjacent weld pairs in the sequence. Weld pairs with higher coupling strengths in this sequence incur a higher cost if they are welded together, meaning concentrated heat input and a greater risk of deformation. For example, in the sequence [W1,W2,W3,W4,W5], the adjacent pairs have a coupling of (W1,W2) = 0.45, (W2,W3) = 0.35, (W3,W4) = 0.40, and (W4,W5) = 0.50, resulting in a total cost of 1.70. In contrast, in the sequence [W1,W4,W2,W5,W3], the adjacent pairs have a coupling of (W1,W4) = 0.10, (W4,W2) = 0.08, (W2,W5) = 0.12, and (W5,W3) = 0.25, resulting in a total cost of only 0.55, which is clearly superior.

[0121] Genetic algorithms iterate and evolve through selection (retaining low-cost individuals), crossover (exchanging parts of two sequences to generate a new sequence), and mutation (randomly swapping the positions of two weld seams). In each generation, low-cost individuals have a higher probability of reproducing, while high-cost individuals are gradually eliminated. Simultaneously, the algorithm checks whether the newly generated sequence satisfies the constraints that W1 and W2 are earlier than W3 and that W4 and W5 are not adjacent; sequences that do not meet these constraints are repaired or removed.

[0122] After dozens of generations of evolution, the total cost of the sequence in the population continuously decreases, eventually converging to a globally approximate optimal solution, such as [W1, W4, W2, W5, W3]. In this sequence, the strongly coupled W1 and W2 are separated by W4, and the strongly coupled W4 and W5 are not adjacent due to W2 in between. The sum of the coupling strength of all adjacent pairs is minimized, thereby achieving the dispersion of heat input and effectively controlling the overall residual deformation after welding. The sequential solution subunit outputs this optimal sequence as the optimized weld bead sequence to the welding module for execution.

[0123] Furthermore, the intelligent welding system in this application embodiment also includes a welding module, which is used to generate a welding program executable by the welding robot based on the compensated welding parameters, so that the welding robot can perform welding operations.

[0124] In the specific implementation process, the system also includes an obstacle avoidance detection module;

[0125] The obstacle avoidance and detection module is specifically used to collect the spatial distance between the robotic arm of the welding robot and the reserved positions of assembled and unassembled parts in real time during welding operations. When the spatial distance is less than the safety threshold, the robot will automatically adjust its movement trajectory or suspend the operation.

[0126] Specifically, the obstacle avoidance and detection module uses laser or vision sensors installed at the end effector and joints of the robotic arm to collect real-time spatial distances between the robotic arm and pre-installed and uninstalled parts at a frequency of at least 50Hz. The system sets multiple safety thresholds: when the distance is greater than 50mm, it only monitors; when entering the 30-50mm range, it automatically decelerates to 50% and issues a warning; when the distance is less than 30mm, it immediately suspends operation and sounds an alarm. Simultaneously, the module has predictive obstacle avoidance capabilities, predicting the minimum distance within 0.5 seconds based on the planned trajectory. If the predicted distance is below the safety threshold, it actively fine-tunes the welding torch posture or local movement path to avoid collisions; if the adjustment is ineffective, it suspends operation. All data is recorded in a log for subsequent analysis, effectively ensuring the safety of the welding robot in complex assembly environments.

[0127] In addition, the system of this application also includes a model update module, which includes:

[0128] The storage unit is used to store the 3D model, deviation vector field, list of weldable welds and compensated welding parameters of each batch of welding operations into the database.

[0129] The training unit is used to periodically retrain the reachability prediction neural network and welding deformation prediction model using historical data in the database, so that the welding accuracy of the system for subsequent batches of hydraulic supports of the same model can be gradually improved.

[0130] The intelligent welding system for hydraulic support structures in this application uses a scanning modeling module to perform high-precision 3D reconstruction of the welding area after each batch of assembly, generating a 3D model reflecting the actual assembly state. A recognition and comparison module locally registers the measured model with the design drawing model, generating a continuous deviation vector field and intelligently distinguishing between weldable and temporarily non-weldable welds, achieving precise quantification of assembly errors and automatic judgment of weld accessibility. The trajectory compensation unit in the adaptive compensation module performs dynamic deformation pre-compensation for weldable welds based on a welding deformation prediction model, generating a compensated 3D welding trajectory. Simultaneously, a sequence optimization unit uses a genetic algorithm with thermal influence coupling strength as weight to solve for the globally optimal weld sequence, achieving dual intelligent optimization of welding trajectory and sequence. The welding module automatically generates a welding program executable by the welding robot, and combined with an obstacle avoidance detection module, it monitors the spatial distance between the robotic arm and the surrounding environment in real time, achieving safe and reliable automated welding operations. The model update module uses historical data to periodically retrain the neural network, continuously improving the system's accuracy. Compared with existing welding methods that rely on manual teaching, fixed trajectories, and cannot adapt to assembly errors, this invention effectively avoids welding defects and collision risks caused by assembly deviations and path obstruction, significantly reduces manual intervention and rework costs, and significantly improves the quality consistency, production efficiency, and intelligence level of hydraulic support structural component welding.

[0131] In the description of this invention, it should be understood that the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0132] In this invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," "linking," and "fixing," etc., should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral part; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication of two components or the interaction between two components. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0133] In this invention, unless otherwise explicitly specified and limited, "above" or "below" the second feature can mean that the first and second features are in direct contact, or that they are in indirect contact through an intermediate medium. Furthermore, "above," "over," or "on top" the second feature can mean that the first feature is directly above or diagonally above the second feature, or simply indicates that the first feature is at a higher horizontal level than the second feature. "Below," "below," or "beneath" the second feature can mean that the first feature is directly below or diagonally below the second feature, or simply indicates that the first feature is at a lower horizontal level than the second feature.

[0134] In the description of this specification, the terms "one embodiment," "some embodiments," "embodiment," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0135] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make modifications, alterations, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. An intelligent welding system for hydraulic support structural components, characterized in that, Based on the welding robot for hydraulic support structural components, it also includes: The scanning modeling module is used to scan the current assembly area of ​​the hydraulic support to be welded after each batch of parts is assembled, and generate a 3D model. The identification and comparison module is used to perform local registration and comparison between the three-dimensional model and the pre-stored design drawing model, obtain the deviation vector field characterizing the current assembly error, and generate a list of deviation items; the list of deviation items includes weldable welds with welding space under the current assembly area and welds that are temporarily unweldable due to adjacent parts not being assembled or path obstruction. An adaptive compensation module is used to perform dynamic deformation pre-compensation and weld sequence optimization on the weldable weld seam based on the deviation vector field and the deviation item list, and generate compensated welding parameters; the welding parameters include the compensated three-dimensional welding trajectory inside each weldable weld seam and the optimized weld sequence between weldable weld seams; The welding module is used to generate a welding program that can be executed by the welding robot based on the compensated welding parameters, so that the welding robot can perform welding operations.

2. The intelligent welding system for hydraulic support structural components according to claim 1, characterized in that, The identification and comparison module specifically includes: The model acquisition unit is used to acquire the theoretical model of the current assembly area corresponding to the three-dimensional model from the pre-stored design drawings based on the three-dimensional model. The deviation calculation unit is used to perform rigid body registration between the three-dimensional model and the theoretical model using the iterative nearest point algorithm, establish a point-to-point correspondence between the three-dimensional model and the theoretical model, and obtain the coordinate difference and normal vector angle of each pair of corresponding points; based on the coordinate difference and normal vector angle of each pair of corresponding points, the three-dimensional position deviation and normal vector angle deviation of each pair of corresponding points are obtained. The field generation unit is used to take the three-dimensional position deviation and normal vector angle deviation of each point as known sample points for radial basis function interpolation, obtain the continuous deviation covering the current assembly area through radial basis function interpolation, and generate the deviation vector field.

3. The intelligent welding system for hydraulic support structural components according to claim 2, characterized in that, The identification and comparison module further includes: The assembly dependency judgment unit is used to mark the welds in the current assembly area that cannot be welded because the adjacent parts have not yet been assembled as assembly-unwelded welds, according to the pre-set batch welding sequence plan. The spatial accessibility determination unit is used to determine whether the weld has welding space based on the three-dimensional model and the deviation vector field; if there is no welding space, it is marked as a weld that cannot be welded temporarily. The list generation unit is used to combine the assembly-temporarily unweldable welds and the space-temporarily unweldable welds into a list of temporarily unweldable welds, and to include the remaining welds that have welding space and no assembly dependency into a list of weldable welds.

4. The intelligent welding system for hydraulic support structural components according to claim 3, characterized in that, The spatial reachability determination unit includes: The information extraction subunit is used to obtain weld information for each weld seam based on the three-dimensional model and the deviation vector field; the weld seam information includes the weld seam's position coordinates, normal offset, and groove geometric parameters; The interference volume ratio acquisition subunit is used to obtain the interference volume ratio when the welding torch approaches each weld seam in the optimal posture using a ray projection algorithm. The interference volume ratio is the percentage of the volume of the welding torch motion envelope intersecting with the surrounding obstacles to the total volume of the welding torch motion envelope. The confidence assessment subunit is used to input the interference volume ratio and weld information of each weld into a pre-trained accessibility prediction neural network to obtain the accessibility confidence of each weld. When the accessibility confidence is greater than a preset threshold, it is determined that there is space for welding.

5. The intelligent welding system for hydraulic support structural components according to claim 4, characterized in that, In the interference volume ratio acquisition subunit, the optimal posture is: Within the preset ranges of the welding torch working angle and travel angle, the welding torch posture is the one that minimizes the ratio of the interference volume between the welding torch motion envelope and surrounding obstacles.

6. The intelligent welding system for hydraulic support structural components according to claim 1, characterized in that, The adaptive compensation module includes: The trajectory compensation unit is used to obtain the dynamic deformation amount of the weldable welds in the deviation item list according to the deviation vector field using the welding deformation prediction model, and to apply the dynamic deformation amount in reverse on the theoretical welding trajectory to generate the compensated three-dimensional welding trajectory inside each weldable weld. The sequence optimization unit is used to solve the globally optimal weld sequence that minimizes the overall residual deformation by using a graph optimization algorithm, based on the process constraints of batch welding of hydraulic supports and with the heat-affected coupling strength between welds as the weight, according to the process constraints of batch welding of hydraulic supports, and to generate the optimized weld sequence.

7. The intelligent welding system for hydraulic support structural components according to claim 6, characterized in that, The welding deformation prediction model in the trajectory compensation unit is a neural network model trained based on historical welding measurement data. The neural network model includes: The input layer is used to receive the local gap of each weldable weld and the preset heat input amount; the local gap is obtained based on the three-dimensional model; The hidden layer consists of at least three fully connected layers and employs ReLU activation and Dropout regularization. The output layer is used to output the dynamic deformation evolution curve of the weldable weld during and after welding. The curve is plotted with time on the horizontal axis and the distribution of dynamic deformation along the weld length on the vertical axis.

8. The intelligent welding system for hydraulic support structural components according to claim 6, characterized in that, The sequence optimization unit includes: The coupling strength acquisition subunit is used to acquire the spatial distance, preset heat input and preset heat dissipation conditions between every two welds in the weldable weld according to the three-dimensional model, and to acquire the heat-affected coupling strength between every two welds according to the spatial distance, preset heat input and preset heat dissipation conditions. The graph construction sub-unit is used to construct a weighted weld sequence relationship graph with each weldable weld as a node and the heat-affected coupling strength as the edge weight. The sequential solution sub-unit is used to employ a genetic algorithm to minimize the overall residual deformation during and after batch welding. Under the premise of satisfying the process constraints of batch welding, the globally optimal traversal order of the weighted weld bead sequence relationship graph is obtained as the optimized weld bead sequence.

9. The intelligent welding system for hydraulic support structural components according to claim 1, characterized in that, The system also includes an obstacle avoidance detection module; The obstacle avoidance detection module is specifically used to collect the spatial distance between the robotic arm of the welding robot and the reserved positions of assembled and unassembled parts in real time during the welding operation. When the spatial distance is less than the safety threshold, the robot automatically adjusts its movement trajectory or suspends the operation.

10. The intelligent welding system for hydraulic support structural components according to claim 1, characterized in that, The system also includes a model update module, which comprises: The storage unit is used to store the 3D model, deviation vector field, list of weldable welds and compensated welding parameters of each batch of welding operations into the database. The training unit is used to periodically retrain the reachability prediction neural network and the welding deformation prediction model using historical data in the database, so that the welding accuracy of the system for subsequent batches of hydraulic supports of the same model can be gradually improved.

Images