A robot adaptive cutting and beveling system for steel structure non-standard nodes

Abstract

The application discloses a kind of robot adaptive cutting and bevel processing systems for steel structure non-standard node, including environment perception module, strategy generation module and intelligent execution module;The environment perception module constructs the dynamic situation model of processing process by fusing vision, force sense and acoustic signal, realizes the deep understanding and state prediction of processing environment;The strategy generation module is based on neural-symbolic computing framework, analyzes the causal relationship between process parameters, generates adaptive processing strategy with explainability;The intelligent execution module is based on the online imitation learning framework of dynamic motion primitive, converts perception information into control instruction, realizes the real-time generation and dynamic disturbance suppression of processing trajectory;The application builds the autonomous cognitive closed loop of perception, decision and execution, so that the processing system has the ability of active adaptation and continuous evolution, effectively solves the technical bottleneck of traditional system in environment understanding, decision transparency and execution adaptability.

Description

Technical Field

[0001] This invention belongs to the field of robot intelligent manufacturing and automated processing technology, specifically relating to a robot adaptive cutting and beveling system for non-standard nodes of steel structures. Background Technology

[0002] Non-standard steel structure nodes are key connection and force transmission components in large and complex spatial steel structure systems. The quality of their cutting and beveling directly affects the connection strength and service safety of the overall structure. As architectural forms increasingly pursue artistry and functionality, the spatial configuration of such nodes is becoming more and more complex and varied, placing extremely high demands on the adaptability and precision of robotic processing technology. Traditional teaching programming and offline trajectory planning methods rely heavily on preset models and human experience, making it difficult to cope with the geometric deviations caused by actual components due to processing errors, clamping postures, and welding deformation, resulting in loss of control over cutting trajectory and beveling accuracy. While existing automated cutting systems have incorporated visual sensing technology, they still suffer from significant limitations in environmental perception under complex working conditions. Specifically, they lack effective filtering and compensation mechanisms for interference from steel structure surface corrosion and cutting fumes; multi-source sensor data is difficult to fuse with high precision, making it impossible to construct dynamic models that reflect the true geometry and processing status of non-standard nodes. Sensing systems are often limited to static dimensional measurements and cannot track micro-deformation and thermal deformation of the structure in real time during processing, leading to inaccurate processing benchmarks and quality fluctuations. At the processing decision-making and execution level, existing methods generally suffer from rigidity and lag in technical bottlenecks. Due to the lack of support from process knowledge bases and digital twins, the system struggles to proactively optimize and adaptively adjust the cutting trajectory and process parameters based on real-time working conditions. The actuators typically follow fixed instructions and cannot achieve smooth interaction with the workpiece under force-position hybrid control, nor do they possess the ability to learn online and control faults in response to sudden working conditions. This rigid control mode severely restricts the improvement of processing efficiency and finished product qualification rate, making it difficult to meet the technical requirements of intelligent construction for high-quality, adaptive processing of complex components. Summary of the Invention

[0003] The purpose of this invention is to provide a robotic adaptive cutting and beveling system for non-standard nodes in steel structures. Based on digital twin and multimodal perception technologies, it realizes real-time perception of the geometry and processing status of non-standard nodes, forward-looking simulation and process optimization of the cutting process, and adaptive control of robot force-position hybrid. It effectively solves the technical problems of perception distortion, decision lag and rigid execution in traditional processing methods, and significantly improves the accuracy, adaptability and overall operation efficiency of non-standard node cutting and beveling.

[0004] The objective of this invention can be achieved through the following technical solutions: This application provides a robotic adaptive cutting and beveling system for non-standard nodes in steel structures, including an environment perception module, a strategy generation module, and an intelligent execution module: Furthermore, the environmental perception module collects the spatial pose, bevel size, and processing status data of non-standard nodes in real time through a multimodal sensor array deployed in the processing area, and constructs a dynamic processing situation model of the non-standard nodes based on an adaptive weighted fusion algorithm; at the same time, the dynamic processing situation model is uploaded to the strategy generation module. Furthermore, the strategy generation module is used to receive the dynamic processing situation model, and to construct and dynamically update a digital twin of the processing process by calling a pre-set non-standard steel structure node process knowledge base through data-driven and model-driven methods; based on the digital twin, to perform forward-looking simulation and optimization of the cutting and beveling process of the non-standard node, generating the optimal cutting trajectory and beveling process parameters; and to generate corresponding adaptive control commands through the optimal cutting trajectory and beveling process parameters and send them to the intelligent execution module. Furthermore, the intelligent execution module is used to execute adaptive control commands from the strategy generation module, and to adjust and optimize the adaptive control commands in real time during the processing based on online learning of dynamic motion primitives and adaptive impedance control, so as to complete the adaptive cutting and beveling of non-standard steel structure nodes. The intelligent execution module feeds back the adaptive cutting and beveling execution status data during the processing to the strategy generation module; the strategy generation module dynamically optimizes the digital twin based on the execution status data and adaptively adjusts the subsequent optimal cutting trajectory and beveling process parameters. Furthermore, the environment perception module includes a point cloud registration unit and an adaptive weighted fusion unit; The point cloud registration unit is configured to perform point cloud registration processing on the spatial pose data collected by the multimodal sensor array based on the improved ICP registration algorithm. Through point cloud denoising, data completion, and multi-view registration, a three-dimensional geometric model of non-standard nodes is constructed. The adaptive weighted fusion unit is configured to perform multi-source information fusion on the three-dimensional geometric model, bevel size data, and processing status data based on the adaptive weighted fusion algorithm to eliminate data distortion caused by steel structure surface corrosion and processing dust through dynamic weight allocation, and to construct a dynamic processing status model.

[0005] Furthermore, the strategy generation module includes a process knowledge base invocation unit, a digital twin construction unit, a multi-objective optimization unit, and a multi-robot collaboration unit; the process knowledge base invocation unit is configured to invoke a pre-set steel structure non-standard node process knowledge base through data-driven and model-driven methods, and based on the dynamic processing situation model, provide process constraints including bevel angle, cutting speed, and gas pressure threshold, as well as historical process data support for the construction of the digital twin; Furthermore, the digital twin construction unit is configured to receive process constraints and historical process data from the process knowledge base calling unit, and construct and dynamically update a digital twin of the processing process using finite element analysis in conjunction with the dynamic processing situation model; the multi-objective optimization unit is configured to perform forward simulation of the cutting and beveling process of non-standard nodes based on the digital twin using a multi-objective optimization algorithm, simultaneously optimizing the cutting trajectory length, beveling angle deviation, and heat-affected zone width, and generating adaptive control instructions containing the optimal cutting trajectory and beveling process parameters; the multi-robot collaboration unit is configured to perform optimal allocation and collaborative trajectory planning of the multi-robot processing area using an improved Hungarian algorithm based on the adaptive control instructions and the digital twin.

[0006] Furthermore, the intelligent execution module includes a trajectory correction unit, an impedance control unit, a quality detection feedback unit, and a fault diagnosis and fault tolerance control unit; the trajectory correction unit is configured to receive structural displacement and stress change data monitored in real time by the environmental perception module based on the adaptive control command; and to learn and adjust the robot's motion trajectory online through dynamic motion primitives, thereby compensating in real time for the relative pose deviation between the robot end and the steel structure caused by workpiece structural deformation due to cutting heat input; Furthermore, the impedance control unit is configured to dynamically adjust the robot's stiffness and damping parameters based on the deviation between the expected and actual contact forces. Through force-position hybrid control, it adaptively adjusts the cutting depth and speed during processing to achieve stable and smooth cutting contact. The quality detection feedback unit is configured to monitor and identify the quality status of the cut surface and the forming status of the cut in real time during the cutting process. Specifically, it uses visual and force sensing fusion to determine whether the cutting is complete, whether there are overheating or slag defects affecting the forming quality of the cut surface, and feeds back the quality status of the cutting process as key execution status data to the strategy generation module and the fault diagnosis and fault tolerance control unit in real time. The fault diagnosis and fault tolerance control unit is configured to diagnose system faults caused during the cutting process in real time based on the execution status data, and trigger a tiered fault tolerance strategy to ensure processing safety and quality when a fault is diagnosed. The tiered fault tolerance strategy includes locking the current processing state, adjusting the processing sequence of multiple robots, and initiating a safety shutdown procedure.

[0007] The beneficial effects of this invention are as follows: This invention achieves adaptive fusion and accurate analysis of multi-source heterogeneous data such as spatial pose, bevel size, and processing status of non-standard nodes through the collaboration of the environmental perception module and the strategy generation module, and constructs a dynamic model that can reflect the processing status in real time. Through the improved point cloud registration algorithm and weighted fusion mechanism, it effectively overcomes the data distortion problem caused by steel structure surface corrosion and processing dust, solves the defect of insufficient perception accuracy of existing technology under complex working conditions, and significantly improves the system's recognition accuracy of the actual geometric features of non-standard nodes and the real-time update of status. By constructing a digital twin that integrates a process knowledge base and combining multi-objective optimization algorithms, the forward-looking simulation and dynamic parameter optimization of the cutting and beveling process are realized. Compared with traditional empirical methods, the finite element thermodynamic analysis and multi-objective Pareto optimization adopted in this invention can simultaneously optimize the cutting trajectory, beveling angle and heat-affected zone range, thereby providing a comprehensive and accurate simulation basis for process parameter decision-making and trajectory planning. This enables the system to predict and avoid quality risks before processing and generate the optimal processing scheme that balances efficiency and quality. This invention combines dynamic motion primitives with adaptive impedance control to achieve online learning and force-position hybrid compliant control of the robot during the processing. By constructing a closed-loop execution architecture that includes real-time trajectory correction, dynamic stiffness and damping adjustment, and quality status feedback, the system can actively compensate for pose deviations caused by structural deformation and immediately trigger fault-tolerant strategies when abnormal contact forces or processing defects are detected. This adaptive control mechanism, which integrates perception, decision-making, execution, and feedback, not only improves the system's adaptability and robustness to dynamic working conditions but also ensures the final quality and process safety of non-standard node cutting and beveling. Attached Figure Description

[0008] To better understand and implement this application, the technical solution is described in detail below with reference to the accompanying drawings.

[0009] Figure 1 A flowchart illustrating a robotic adaptive cutting and beveling system for non-standard steel structure nodes provided in this application; Figure 2 A schematic diagram illustrating the process of generating and updating a digital twin model for a robot adaptive cutting and beveling system for non-standard steel structure nodes provided in this application; Detailed Implementation

[0010] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, exemplary embodiments will be described in detail below, examples of which are illustrated in the accompanying drawings. In the following description relating to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of methods and systems consistent with some aspects of this application as detailed in the appended claims.

[0011] The terminology used in this application is for the purpose of describing particular embodiments only and is not intended to be limiting of the application. The singular forms “a,” “the,” and “the” used in this application and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used herein refers to and includes any or all possible combinations of one or more of the associated listed items.

[0012] The following detailed description of the specific implementation methods, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided in detail.

[0013] Example 1 Please see Figures 1-2 This embodiment provides a robotic adaptive cutting and beveling system for non-standard nodes in steel structures, including an environmental perception module, a strategy generation module, and an intelligent execution module.

[0014] The environmental perception module uses a multimodal sensor array deployed in the processing area to collect real-time data on the spatial pose, bevel dimensions, and processing status of non-standard nodes. Based on an adaptive weighted fusion algorithm, it constructs a dynamic processing status model of the non-standard nodes and uploads this model to the strategy generation module. The strategy generation module receives the dynamic processing status model and, through data-driven and model-driven approaches, calls a pre-set steel structure non-standard node process knowledge base to construct and dynamically update a digital twin of the processing process. Based on this digital twin, it performs forward-looking simulation and optimization of the cutting and beveling process of the non-standard nodes, generating optimal cutting trajectories, beveling process parameters, and corresponding adaptive control commands, which are then sent to the intelligent execution module. The intelligent execution module executes the adaptive control commands from the strategy generation module and, based on online learning of dynamic motion primitives and adaptive impedance control, adjusts and optimizes the commands in real-time during the processing, completing the adaptive cutting and beveling of the steel structure non-standard nodes. The intelligent execution module feeds back the adaptive cutting and beveling execution status data during the processing to the strategy generation module, forming a complete closed-loop control system. Specifically, this embodiment solves the technical problems of distorted state perception, decision-making dependence on experience, and rigid execution in traditional processing methods by constructing a closed-loop control architecture of perception, decision-making, execution, and optimization. It achieves the technical effects of accurate processing trajectory of non-standard nodes, consistent bevel quality, and strong system adaptability. Furthermore, the environment perception module includes a point cloud registration unit and an adaptive weighted fusion unit; The point cloud registration unit performs point cloud registration processing on the spatial pose data collected by the multimodal sensor array based on the improved ICP registration algorithm. Through point cloud denoising, data completion and multi-view registration, a three-dimensional geometric model of non-standard nodes is constructed. Specifically, the point cloud registration process for spatial pose data acquired by the multimodal sensor array based on the improved ICP registration algorithm is as follows: Outlier removal and voxel mesh downsampling are performed on the initial point cloud using statistical filtering, with the filtering threshold set to 2.5 times the standard deviation and the voxel size set to 2 mm; the missing regions of the point cloud are reconstructed and data is completed using the Poisson reconstruction algorithm, with a reconstruction depth set to 9; during the registration iteration process, the rate of decrease of the current point cloud alignment error is calculated in real time; when the point cloud registration unit detects that the error decrease rate is lower than the set sensitivity threshold multiple times consecutively, it determines that the convergence process has entered a plateau period. At this time, the iteration step size is automatically reduced proportionally, causing the ICP registration algorithm to switch to a fine convergence mode, avoiding oscillations caused by excessively large step sizes. Simultaneously, the ICP registration algorithm analyzes the point cloud distribution characteristics of the current processing area: in sparse point cloud areas, to ensure matching success rate, the point cloud registration unit expands the search range of nearest neighbor points proportionally; in dense point cloud areas, to improve computational efficiency and reduce false matching, the point cloud registration unit correspondingly narrows the search range; the threshold for determining sparse and dense point clouds and the scaling ratio of the search range are pre-calibrated through statistical analysis of typical steel structure point cloud samples and called in real time during the registration process, thereby realizing dynamic optimization of feature matching strategies for different regions of complex structure surfaces; Furthermore, the adaptive weighted fusion unit, based on the adaptive weighted fusion algorithm, eliminates data distortion caused by steel structure surface corrosion and processing dust through dynamic weight allocation, and performs multi-source information fusion on the three-dimensional geometric model, bevel size data, and processing status data to construct a dynamic processing situation model; Specifically, the implementation process of eliminating data distortion caused by steel structure surface corrosion and processing dust through dynamic weight allocation based on the adaptive weighted fusion algorithm is as follows: Quality evaluation indicators for the three-dimensional geometric model, bevel dimension data, and processing status data are calculated respectively. The integrity of point cloud data in the corroded area, the measurement stability of bevel dimensions under dust interference, and the signal-to-noise ratio of processing status load data are key evaluation indicators. Based on the evaluation indicators, fusion weight coefficients are dynamically allocated to the three-dimensional geometric model, bevel dimension data, and processing status data to form a weight allocation strategy. Regions with missing point clouds due to corrosion are assigned a geometric weight of 0.2-0.4, bevel dimension measurement data under dust interference are assigned a size weight of 0.3-0.5, and stable processing load data collected by force sensors are assigned a high weight of 0.7-0.9. Based on the results of the dynamic weight allocation, a Kalman filter is used for multi-source information fusion. The process noise covariance is set to 0.01, and the observation noise covariance is set to 0.1. After 5-8 iterations, a dynamic processing status model is generated. Furthermore, the environmental perception module integrates an improved ICP registration algorithm with an adaptive weighted fusion algorithm based on quality evaluation to construct a perception architecture in which the point cloud registration unit and the adaptive weighted fusion unit work together. This effectively solves the core technical problems of missing 3D point cloud data, noise interference, and conflicts of multi-source sensor information caused by steel structure surface corrosion, processing dust, and complex spatial structure occlusion. As a result, it achieves the technical effect of real-time construction and output of high-precision, high-completeness, and high-reliability non-standard node dynamic processing situation model, providing a precise and stable perception data foundation for subsequent strategy generation and intelligent execution. Furthermore, the strategy generation module includes a process knowledge base calling unit, a digital twin construction unit, a multi-objective optimization unit, and a multi-robot collaboration unit.

[0015] The process knowledge base calling unit calls the pre-set process knowledge base for non-standard steel structure nodes to provide process constraints for the construction of digital twins, such as bevel angle of 30°-60°, cutting speed of 0.8-1.5m / min, and gas pressure of 0.6-0.8MPa. The digital twin construction unit is based on a dynamic processing situation model, combined with process constraints, and uses finite element analysis to construct and dynamically update the digital twin of the processing process; Q345 steel parameters are used: thermal conductivity 42W / (m·K), specific heat capacity 480J / (kg·K), density 7850kg / m³. 3 With the mesh size set to 1.5 mm, the temperature field distribution and thermal deformation during the cutting process were simulated through transient thermo-mechanical coupling analysis. The multi-objective optimization unit is based on a digital twin and uses a multi-objective optimization algorithm to perform a forward-looking simulation of the cutting and beveling process, simultaneously optimizing the cutting trajectory length, beveling angle deviation, and heat-affected zone width, and generating adaptive control commands that include the optimal cutting trajectory and beveling process parameters. Specifically, the implementation process of using a multi-objective optimization algorithm to perform a forward-looking simulation of the cutting and beveling process is as follows: By establishing an objective function for the length of the steel structure cutting trajectory, the shortest processing path and bevel angle deviation are calculated based on the actual three-dimensional geometric features of non-standard nodes. The objective function sets an optimization range of 30° to 60° for steel structure nodes with different plate thicknesses. The objective function for the width of the heat-affected zone is quantified by combining the thermal conductivity, specific heat capacity and density parameters of the steel through a finite element heat conduction model. In the digital twin, based on the quantification results, the temperature field distribution during the movement of the cutting heat source is simulated through transient thermodynamic analysis, and the range of the heat-affected area and thermal deformation under different combinations of cutting speed and gas pressure are predicted. A multi-objective optimization algorithm is adopted, and the diversity of the solution set is maintained through the reference point selection mechanism. The optimization strategy is adaptively adjusted for steel structures with different plate thicknesses. The Pareto optimal solution with excellent cutting efficiency, bevel quality and thermal damage control is selected through non-dominated sorting. The best combination of process parameters with comprehensive performance is selected in combination with the steel structure welding process specifications and fed back to the digital twin for verification simulation. The multi-robot collaborative unit is based on adaptive control commands and digital twins. It uses an improved Hungarian algorithm to optimally allocate the processing area of ​​the multi-robots and plan the collaborative trajectory. It constructs a cost matrix that considers the robot's motion state, load capacity and task position, introduces dynamic weighting factors and conflict avoidance mechanisms, and uses a collaborative trajectory planning algorithm with fused time windows to generate collision-free and smooth motion trajectories. Specifically, the implementation process of optimal allocation and collaborative trajectory planning of multi-robot processing areas using the improved Hungarian algorithm is as follows: Construct a cost matrix for multi-robot task allocation; each element of the cost matrix represents the comprehensive cost of the robot completing a sub-processing task. The factors for calculating the comprehensive cost include: the Euclidean distance of the robot end effector moving from its current position to the task start point, the energy consumption coefficient corresponding to the current joint load state of the robot, the matching degree between the processing accuracy required by the task and the robot's repeatability accuracy, and the preset process priority weight of the task. A dynamic conflict avoidance factor is introduced into the cost matrix. Before each iteration of allocation calculation, based on the real-time position of each robot and the spatial position relationship of the task, potential trajectory intersection areas are predicted, and a penalty term cost is added to the cost matrix for task pairs involving potential collision risks, guiding the algorithm to automatically avoid spatial conflicts. An improved Hungarian algorithm is used to solve the cost matrix after introducing a dynamic conflict avoidance factor. During the row and column subtraction and zero-element marking process of the standard Hungarian algorithm, a priority elimination mechanism for the penalty cost is added to ensure that the algorithm prioritizes the allocation of collision-free or low-collision-risk task combinations during optimization. The algorithm outputs a one-to-one correspondence between robots and tasks that minimizes the total comprehensive cost, serving as the optimal processing task allocation scheme for collaborative trajectory planning for each robot. A path search algorithm with fused time windows is used to generate a collision-free path for each robot from the starting point to the task point, and then to the start and end points of the processing trajectory. The time window is dynamically defined by estimating the standard running time of each robot in each path segment and superimposing an adaptive time margin based on its load state. During planning, it is ensured that the time windows occupied by each robot in the shared space do not overlap, thereby avoiding collisions from a temporal perspective. The generated initial path is smoothed and optimized; B-spline curves are used to fit the path nodes, and the robot joint acceleration and jerk are constrained to be continuous. Under the premise of satisfying the no-collision and time window constraints, a smooth motion trajectory with high efficiency and low vibration is generated.

[0016] Furthermore, the strategy generation module constructs and dynamically updates a digital twin, and integrates multi-objective optimization algorithms and improved Hungarian algorithms for forward-looking simulation and collaborative planning. This solves the key technical problems in traditional processing, such as rigid process decisions, inability to adapt to dynamic working conditions in real time, and conflicts and low efficiency of multiple robot tasks caused by relying on fixed programs and human experience. It achieves the technical effect of generating and dynamically optimizing the cutting trajectory, process parameters and conflict-free collaborative instructions with the best comprehensive performance in real time for the execution layer. Specifically, the strategy generation module, driven by both data and models, integrates real-time sensing data with a process knowledge base to construct a high-fidelity dynamic digital twin, solving the problem of asynchrony between the physical world and the virtual model. Based on this, a multi-objective optimization algorithm is used to simultaneously simulate and optimize the cutting trajectory, bevel quality, and thermal effects, generating a Pareto-optimal process solution, replacing experience-based decisions relying on a single objective. Simultaneously, an improved Hungarian algorithm is used for optimal allocation of multi-robot tasks and collaborative trajectory planning, resolving issues of load imbalance and motion interference in multi-machine processing. Ultimately, this achieves an intelligent upgrade of the processing solution from static pre-setting to dynamic optimization, and from single-machine execution to multi-machine collaboration, enabling the system to automatically adapt to complex and ever-changing non-standard node processing scenarios. The intelligent execution module includes a trajectory correction unit, an impedance control unit, a quality detection feedback unit, and a fault diagnosis and fault tolerance control unit; The trajectory correction unit, based on the adaptive control command and receiving data monitored in real time by the environmental perception module, learns and adjusts the robot's motion trajectory online through dynamic motion primitives, and compensates in real time for the relative pose deviation between the robot end and the steel structure caused by workpiece structural deformation due to cutting heat input. The impedance control unit dynamically adjusts the robot's stiffness and damping parameters based on the deviation between the expected and actual contact forces. Adaptive adjustment is performed during processing via force-position hybrid control. Based on the digital twin, the adaptive impedance controller dynamically adjusts the stiffness and damping parameters of the robot's end effector according to the deviation between the expected contact force and the actual contact force measured by the six-dimensional force sensor. When the actual contact force is more than 10% greater than the expected value, the stiffness parameter is reduced from 5000 N / m to 3000 N / m, and the damping parameter is increased from 200 N·s / m to 300 N·s / m. When the actual contact force is more than 10% less than the expected value, the stiffness parameter is increased to 6000 N / m. During the cutting process, the quality inspection feedback unit monitors the quality status of the cutting surface and the cutting edge formation status in real time through the fusion of visual and force sensing. It judges whether the cutting is thorough, whether there are overheating or slag defects, and feeds the processing quality status data back to the strategy generation module and the fault diagnosis and fault tolerance control unit in real time. The fault diagnosis and fault tolerance control unit diagnoses system faults in the cutting process in real time based on execution status data, and triggers a hierarchical fault tolerance strategy when a fault is diagnosed: when abnormal force sensor data is detected, the current processing state is locked; when conflicts occur in multi-robot collaboration, the processing sequence is adjusted; when a serious fault occurs, a safety shutdown procedure is initiated. Specifically, the process of diagnosing system faults in the cutting process in real time based on execution status data, and triggering the implementation of a hierarchical fault-tolerance strategy when a fault is diagnosed, is as follows: For transient interference and recoverable anomalies: When the fault diagnosis and fault tolerance control unit detects transient fluctuations in data such as contact force, image features, or trajectory tracking deviation reported by the quality detection feedback unit or trajectory correction unit, but these fluctuations do not exceed the process allowable range set by the digital twin, they are determined to be recoverable transient interference. At this time, the fault diagnosis and fault tolerance control unit immediately issues a command to the trajectory correction unit and the process system to lock the current processing state: the trajectory correction unit controls the robot to pause its movement and precisely maintain the end-effector pose, while the system shuts off the cutting heat source; subsequently, the fault diagnosis and fault tolerance control unit sends the current state and the latest data from the environmental perception module to the strategy generation module. Based on this, the strategy generation module performs online simulation and fine-tuning of the process parameters (such as cutting speed and gas flow rate) at the current pose point in its digital twin. If, within a set short delay (such as 500ms), the feedback data from the environmental perception module returns to normal and the simulation verification passes, the strategy generation module sends out the optimized parameters, the fault diagnosis and fault tolerance control unit immediately unlocks, and the instruction system automatically resumes processing from the pause point using the new parameters; For persistent local faults and collaborative conflicts: When the fault diagnosis and fault tolerance control unit diagnoses persistent local faults such as continuous abnormal sensor data, inability to identify local quality features, or potential interference in the movement of multiple robots, the system triggers task reconfiguration; the fault diagnosis and fault tolerance control unit sends the fault information to the strategy generation module to adjust the processing order of multiple robots; the strategy generation module, based on the updated digital twin, dynamically reallocates the processing areas of the remaining available robots and replans local trajectories through the multi-robot collaboration unit, generating new conflict-free collaborative instructions, enabling the process system to continue operating in degraded mode. At the same time, the fault diagnosis and fault tolerance control unit synchronizes the state of the process system entering task reconfiguration and the new task layout to the monitoring interface and updates the system log; Emergency safety isolation for severe faults: When the fault diagnosis and fault tolerance control unit diagnoses a severe fault that may endanger equipment or personal safety (such as severe out-of-tolerance motion axis, sudden collision, or communication interruption), the fault diagnosis and fault tolerance control unit immediately sends a safety shutdown initiation procedure to all robot drives and process power supplies through its independent hardware safety loop. The safety shutdown initiation procedure forces all robot drives to power off and enter mechanical brake mode, simultaneously shutting down all process equipment and power sources, putting the process system into a safety lockout state, and triggering audible and visual alarms. The process system will remain in this state until authorized personnel complete the troubleshooting and reset. Specifically, the implementation process, through real-time interaction with the strategy generation module and direct control of independent hardware, achieves a progressive response from parameter self-correction and dynamic task reconstruction to emergency safety isolation, ensuring the adaptive capability and inherent safety of the processing process in the face of different levels of faults. Furthermore, the intelligent execution module solves key technical problems such as execution trajectory deviation, processing instability, quality loss of control and system safety risks caused by workpiece thermal deformation, contact force fluctuation, quality defects and sudden abnormalities during the processing through the closed-loop collaboration of the integrated trajectory correction unit, impedance control unit, quality detection feedback unit and fault diagnosis and fault tolerance control unit. It achieves the technical effect of high-precision, compliant, adaptive and safe and reliable cutting and beveling execution. Specifically, this module compensates for trajectory deviations in real time based on online learning of dynamic motion primitives, achieves force-position hybrid compliant processing based on adaptive impedance control, monitors and provides feedback on quality in real time through multi-sensor fusion, and ensures the safe operation of the system through a hierarchical fault-tolerant strategy. In this way, decision commands are accurately, flexibly and robustly transformed into processing actions in the physical world, ensuring the quality of the final product and the safety and controllability of the entire processing process. This embodiment was implemented in a large-scale transportation hub steel structure project, processing 156 non-standard intersecting nodes; the system was equipped with 6 KUKA KR60 industrial robots, 12 sets of laser scanners, and 18 sets of six-dimensional force sensors. During implementation, the environmental perception module collected data at a frequency of 100Hz, the strategy generation module updated the digital twin at a frequency of 20Hz, and the intelligent execution module had a control cycle of 5ms. The results show that the system achieves a cutting trajectory tracking accuracy of ±0.08mm, a bevel angle deviation of ±0.3°, a heat-affected zone width of 1.8mm, and a processing efficiency that is 35% higher than that of traditional processes. The first-pass yield rate of the products reaches 98.5%, which effectively verifies the technical effect of the system of the present invention.

[0017] Example 2 This embodiment details the application of the robot adaptive cutting and beveling system for non-standard steel structure nodes in the processing of non-standard cast steel nodes in large-span spatial grid structures. Compared with Embodiment 1, the core difference of this embodiment is that, addressing the technical challenges of variable spatial postures of non-standard nodes in large-span spatial structures, limited robot accessibility, and complex thermal deformation due to large differences in structural rigidity, this embodiment solves key technical problems such as poor trajectory planning adaptability, low multi-machine collaboration efficiency, and difficulty in controlling thermal accumulation deformation in traditional robot processing methods in such complex applications by constructing a hierarchical multi-robot collaborative architecture and implementing a priori thermal deformation compensation strategy. The hierarchical multi-robot collaborative architecture is a preferred implementation scheme of the multi-robot collaborative unit; the thermal deformation prior compensation strategy is a deep integration and extended application of the process knowledge base calling, digital twin construction and multi-objective optimization process; The system implementation process specifically includes the following optimization phases: High-precision perception and adaptive modeling stage: In the environmental perception module, considering the complex surface conditions of cast steel nodes (with casting seams and oxide scale) and the susceptibility of spatial pose measurement benchmarks to on-site vibration interference, a multi-view fusion perception network consisting of four high-precision 3D structured light scanners (with repeatability of ±0.02mm) and two laser trackers is deployed at the processing station. The point cloud registration unit adopts an improved ICP algorithm and improves the registration accuracy of multi-view point clouds to ±0.05mm by introducing a global benchmark provided by the trackers. The adaptive weighted fusion unit, considering the characteristics of the cast surface, introduces the surface curvature change rate as a quality evaluation index in the weight allocation strategy, assigning higher weights to point cloud data in high curvature feature areas to ensure accurate extraction of complex contour features and provide high-fidelity geometric input for the construction of digital twins. Hierarchical collaborative planning and dynamic allocation phase: The multi-robot collaborative unit in the strategy generation module, when implementing the improved Hungarian algorithm, adopts the hierarchical collaborative architecture as follows: Multiple robots in the processing unit are divided into main processing robots (equipped with high-power cutting heads) and auxiliary operation robots (equipped with slag removal, measurement, or auxiliary support devices). When allocating tasks, the cost matrix not only considers the robot's motion and load state but also incorporates the processing accessibility index and regional rigid support requirements obtained based on digital twin analysis. The system prioritizes the main robot to undertake core beveling tasks, while for areas with poor accessibility for the main robot or requiring auxiliary support to prevent vibration, the collaborative unit dynamically allocates these tasks to auxiliary robots for auxiliary or collaborative processing operations. This hierarchical collaborative architecture increases the executability rate of processing tasks from approximately 75% using traditional methods to 98% when facing extreme spatial attitude nodes. Thermal deformation prediction and online compensation control stage: The strategy generation module's digital twin construction unit is deeply integrated with the multi-objective optimization unit to implement a priori compensation strategy for thermal deformation. When constructing the digital twin, in addition to general material parameters, the process knowledge base calling unit provides a historical thermal deformation case library for different node plate thickness combinations and joint types. During the prospective simulation, the strategy generation module first performs case matching based on the current node characteristics, preloading the deformation trends of similar cases as initial conditions into the thermo-mechanical coupling simulation, significantly improving the accuracy of the first round of prediction. In the actual processing, the trajectory correction unit of the intelligent execution module not only compensates based on real-time monitoring data but also uses the deviation between the actual thermal deformation data and the predicted value as a real-time compensation learning signal. The strategy generation module uses this signal to dynamically update the material property parameters (such as the coefficient of thermal expansion) related to thermal deformation in the digital twin through an online learning algorithm, continuously improving the model's prediction accuracy during processing. Full-process quality monitoring and collaborative safety phase: The intelligent execution module's quality detection feedback unit and fault diagnosis and tolerance control unit work together. During the cutting process, in addition to monitoring the quality of the cut surface, the quality detection unit also monitors the temperature gradient of key points around the cutting area through a thermal imager mounted on the auxiliary robot, serving as an indirect verification of the thermal deformation state. The fault diagnosis unit is equipped with safety logic based on the consistency of multi-source information: when the compensation amount of the trajectory correction unit continues to increase, and the thermal imager detects an abnormally high temperature area, the fault diagnosis and tolerance control unit determines that there is a potential risk of uncontrolled thermal deformation and immediately triggers the hierarchical fault tolerance strategy. First, the cutting speed of the main robot and the cooling assistance strategy of the auxiliary robot are adjusted; if this is ineffective, the current path is paused, and the strategy generation module replans the subsequent cutting sequence to disperse the heat input; ultimately ensuring processing safety and forming quality. In the processing of a critical cast steel node (main pipe thickness 45mm, branch pipe thickness 25mm, and variable spatial angle) in a large spatial grid structure, the system predicted a high risk of thermal deformation in the thickness transition zone. The multi-objective optimization unit actively reduced the cutting speed by 15% in the risk zone and instructed the auxiliary robot to simultaneously start local air cooling. In actual processing, the quality inspection feedback unit monitored that the deformation in this area was reduced by about 40% compared with similar nodes without compensation measures, and the bevel angle deviation was stable within ±0.4°, verifying the effectiveness of the thermal deformation prior compensation strategy. This embodiment, through the implementation of the hierarchical multi-robot collaborative architecture and the thermal deformation prior compensation strategy, achieves 100% machining accessibility for complex spatial cast steel nodes, reduces the deviation rate of beveling accuracy caused by thermal deformation by more than 60%, and improves overall machining efficiency by 30%. Particularly in the machining of several of the most complex non-standard nodes, the system successfully avoided rework due to thermal accumulation deformation through predictive control, verifying the high adaptability and reliability of the system described in this invention in addressing the challenges of machining high-difficulty nodes in large-span spatial structures.

[0018] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A robotic adaptive cutting and beveling system for non-standard nodes in steel structures, characterized in that: It includes an environmental perception module, a policy generation module, and an intelligent execution module; The environmental perception module collects the spatial pose, bevel size, and processing status data of non-standard nodes in real time through a multimodal sensor array deployed in the processing area, and constructs a dynamic processing situation model of the non-standard nodes based on an adaptive weighted fusion algorithm; at the same time, the dynamic processing situation model is uploaded to the strategy generation module. The strategy generation module is used to receive the dynamic processing situation model, and to construct and dynamically update a digital twin of the processing process by calling a pre-set non-standard steel structure node process knowledge base through data-driven and model-driven methods; based on the digital twin, to perform forward-looking simulation and optimization of the cutting and beveling process of the non-standard node, and to generate the optimal cutting trajectory and beveling process parameters; and to generate corresponding adaptive control commands based on the optimal cutting trajectory and beveling process parameters and send them to the intelligent execution module. The intelligent execution module is used to execute adaptive control commands from the strategy generation module, and to adjust and optimize the adaptive control commands in real time during the processing based on online learning of dynamic motion primitives and adaptive impedance control, so as to complete the adaptive cutting and beveling of non-standard steel structure nodes. The intelligent execution module feeds back the adaptive cutting and beveling execution status data during the processing to the strategy generation module. The strategy generation module dynamically optimizes the digital twin based on the execution status data, and adaptively adjusts the subsequent optimal cutting trajectory and beveling process parameters.

2. The robotic adaptive cutting and beveling system for non-standard nodes of steel structures according to claim 1, characterized in that: The environment perception module includes a point cloud registration unit and an adaptive weighted fusion unit; The point cloud registration unit is configured to perform point cloud registration processing on the spatial pose data collected by the multimodal sensor array based on the improved ICP registration algorithm, and to construct a three-dimensional geometric model of non-standard nodes through point cloud denoising, data completion and multi-view registration. The adaptive weighted fusion unit is configured to eliminate data distortion caused by steel structure surface corrosion and processing dust through dynamic weight allocation based on the adaptive weighted fusion algorithm, and to perform multi-source information fusion on the three-dimensional geometric model, bevel size data and processing status data to construct a dynamic processing situation model.

3. The robotic adaptive cutting and beveling system for non-standard nodes in steel structures according to claim 2, characterized in that, The process of performing point cloud registration processing on the spatial pose data acquired by the multimodal sensor array based on the improved ICP registration algorithm specifically includes: The initial point cloud acquired by the multimodal sensor array is subjected to outlier removal based on statistical filtering and voxel mesh downsampling to eliminate corrosion interference on the steel structure surface; the missing areas of the point cloud are reconstructed and data is completed using the Poisson reconstruction algorithm to repair visual occlusion areas caused by the complex structure of non-standard nodes; an improved ICP registration algorithm is adopted, using the distance from point to surface as the error metric, and the iteration step size is adaptively adjusted based on the point cloud matching error during the registration iteration process of the ICP registration algorithm. At the same time, the nearest neighbor search threshold is adaptively optimized based on the local point cloud density for high-precision registration of multi-view point cloud data. The process of eliminating rust interference on the steel structure surface is as follows: Based on the statistical analysis of the average distance between each point in the point cloud and its nearest neighbor, a dynamic distance threshold is set, and points whose average distance exceeds the dynamic distance threshold are identified as outliers representing surface rust and deposits and are filtered out; based on voxel mesh downsampling of the filtered point cloud, the corresponding surface normal vector is estimated for each data point in the downsampled point cloud, and a normal vector field covering the entire point cloud is constructed; the normal vector field is used as the divergence source of the gradient field to construct a Poisson equation to describe the relationship between the inside and outside of the target surface; the Poisson equation is discretized into a linear system through an octree mesh structure with defined rules in three-dimensional space, and solved by the conjugate gradient method to obtain a continuous scalar indicator function; an isosurface with a specific value is extracted from the scalar indicator function, and the isosurface is transformed into a closed, watertight triangular mesh surface model using the moving cube algorithm; based on the triangular mesh surface model, new point cloud data is generated to fill in the missing areas in the original point cloud caused by visual occlusion.

4. The robotic adaptive cutting and beveling system for non-standard nodes in steel structures according to claim 2, characterized in that, The process of eliminating data distortion caused by steel structure surface corrosion and processing dust through dynamic weight allocation based on the adaptive weighted fusion algorithm specifically includes: The quality evaluation indexes of the three-dimensional geometric model, bevel size data and processing status data are calculated based on the adaptive weighted fusion algorithm. Among them, for the surface characteristics of steel structure, the quality evaluation indexes focus on evaluating the integrity of point cloud data in the rusted area, the measurement stability of bevel size under the interference of smoke and dust, and the signal-to-noise ratio of processing status load data. Based on the results of the key evaluation of the quality evaluation indicators, the fusion weight coefficients are dynamically assigned to the three-dimensional geometric model, bevel size data and processing status data to form a weight allocation strategy: low geometric weight is assigned to the area where the point cloud is missing due to corrosion, low size weight is assigned to the measurement data of bevel size under the interference of smoke and dust, and high weight is assigned to the stable processing load data collected by the force sensor. Based on the weight allocation strategy, a Kalman filter is used to fuse the three-dimensional geometric model, bevel size data, and processing status data. The deviation of point cloud data caused by surface corrosion and processing dust is calculated iteratively and gradually corrected to generate a dynamic processing status model that reflects the actual geometric characteristics and processing status of non-standard nodes.

5. The robotic adaptive cutting and beveling system for non-standard nodes of steel structures according to claim 1, characterized in that: The strategy generation module includes a process knowledge base calling unit, a digital twin construction unit, a multi-objective optimization unit, and a multi-robot collaboration unit; The process knowledge base calling unit is configured to call a pre-set process knowledge base for non-standard steel structure nodes through data-driven and model-driven methods. Based on the dynamic processing situation model, it provides process constraints and historical process data support for the construction of digital twins, including bevel angle, cutting speed, and gas pressure threshold. The digital twin construction unit is configured to receive process constraints and historical process data from the process knowledge base calling unit, and combine the dynamic processing situation model with finite element analysis to construct and dynamically update the digital twin of the processing process. The multi-objective optimization unit is configured to perform a forward-looking simulation of the cutting and beveling process of non-standard nodes based on the digital twin using a multi-objective optimization algorithm, simultaneously optimizing the cutting trajectory length, beveling angle deviation, and heat-affected zone width, and generating adaptive control commands containing the optimal cutting trajectory and beveling process parameters. The multi-robot collaborative unit is configured to perform optimal allocation and collaborative trajectory planning of the multi-robot processing area based on the adaptive control commands and the digital twin using an improved Hungarian algorithm.

6. The robotic adaptive cutting and beveling system for non-standard steel structure nodes according to claim 5, characterized in that, The process of prospectively simulating the cutting and beveling of non-standard nodes using a multi-objective optimization algorithm, and simultaneously optimizing the cutting trajectory length, beveling angle deviation, and heat-affected zone width, specifically includes: By establishing an objective function for the length of the steel structure cutting trajectory, the shortest processing path and bevel angle deviation are calculated based on the actual three-dimensional geometric features of non-standard nodes. The objective function sets an optimization range of 30° to 60° for steel structure nodes with different plate thicknesses. The objective function for the width of the heat-affected zone is quantified by combining the thermal conductivity, specific heat capacity and density parameters of the steel through a finite element heat conduction model. In the digital twin, based on the quantification results, the temperature field distribution during the movement of the cutting heat source is simulated through transient thermodynamic analysis, and the range of the heat-affected area and thermal deformation under different combinations of cutting speed and gas pressure are predicted. A multi-objective optimization algorithm is adopted, and the diversity of the solution set is maintained through a reference point selection mechanism. The optimization strategy is adaptively adjusted for steel structures with different plate thicknesses. The Pareto optimal solution with excellent cutting efficiency, bevel quality and thermal damage control is selected through non-dominated sorting. The combination of process parameters with the best comprehensive performance is selected in combination with the steel structure welding process specifications and fed back to the digital twin for verification simulation.

7. The robotic adaptive cutting and beveling system for non-standard steel structure nodes according to claim 5, characterized in that, An improved Hungarian algorithm is used to perform optimal allocation and cooperative trajectory planning for multi-robot processing areas, specifically including: Construct a cost matrix for multi-robot task allocation; each element of the cost matrix represents the comprehensive cost of the robot completing a sub-processing task. The factors for calculating the comprehensive cost include: the Euclidean distance of the robot end effector moving from its current position to the task start point, the energy consumption coefficient corresponding to the current joint load state of the robot, the matching degree between the processing accuracy required by the task and the robot's repeatability accuracy, and the preset process priority weight of the task. A dynamic conflict avoidance factor is introduced into the cost matrix. Before each iteration of allocation calculation, based on the real-time position of each robot and the spatial position relationship of the task, potential trajectory intersection areas are predicted, and a penalty term cost is added to the cost matrix for task pairs involving potential collision risks, guiding the algorithm to automatically avoid spatial conflicts. An improved Hungarian algorithm is used to solve the cost matrix after introducing a dynamic conflict avoidance factor. During the row and column subtraction and zero element marking process of the standard Hungarian algorithm, a priority elimination mechanism for the cost of the penalty term is added to ensure that the algorithm prioritizes the allocation of task combinations with no collision or low collision risk during optimization. The algorithm outputs a one-to-one correspondence between the robot and the task that minimizes the total comprehensive cost, which serves as the optimal processing task allocation scheme for collaborative trajectory planning for each robot.

8. The robotic adaptive cutting and beveling system for non-standard nodes of steel structures according to claim 7, characterized in that, The optimal processing task allocation scheme involves collaborative trajectory planning for each robot, specifically including: A path search algorithm with fused time windows is used to generate a collision-free path for each robot from the starting point to the task point and then to the start and end points of the processing trajectory. The time window is dynamically defined by estimating the standard running time of each robot in each path segment and superimposing an adaptive time margin based on its load state. During planning, it is ensured that the time windows occupied by each robot in the shared space do not overlap, thereby avoiding collisions in the time dimension. The generated initial path is smoothed and optimized; B-spline curves are used to fit the path nodes, and the robot joint acceleration and jerk are constrained to be continuous. Under the premise of satisfying the no-collision and time window constraints, a smooth motion trajectory with high efficiency and low vibration is generated.

9. The robotic adaptive cutting and beveling system for non-standard nodes in steel structures according to claim 1, characterized in that: The intelligent execution module includes a trajectory correction unit, an impedance control unit, a quality detection feedback unit, and a fault diagnosis and fault tolerance control unit; The trajectory correction unit is configured to receive, based on the adaptive control command, structural displacement and stress change data monitored in real time by the environmental perception module; By learning and adjusting the robot's motion trajectory online through dynamic motion primitives, the relative pose deviation between the robot's end effector and the steel structure caused by workpiece structural deformation due to cutting heat input can be compensated in real time. The impedance control unit is configured to dynamically adjust the stiffness and damping parameters of the robot according to the deviation between the expected and actual contact force, and to adaptively adjust the cutting depth and speed during the processing through force-position hybrid control to achieve stable and smooth cutting contact. The quality inspection feedback unit is configured to monitor and identify the quality status of the cut surface and the forming status of the cut in real time during the cutting process. Specifically, it uses visual and force sensing fusion to determine whether the cutting is complete, whether there are defects such as overheating and slag affecting the forming quality of the cut surface, and feeds back the quality status of the cutting process as key execution status data to the strategy generation module and the fault diagnosis and fault tolerance control unit in real time. The fault diagnosis and fault tolerance control unit is configured to diagnose system faults caused during the cutting process in real time based on the execution status data, and trigger a hierarchical fault tolerance strategy to ensure processing safety and quality when a fault is diagnosed; the hierarchical fault tolerance strategy includes locking the current processing status, adjusting the processing sequence of multiple robots, and starting a safety shutdown procedure.

10. A robotic adaptive cutting and beveling system for non-standard nodes in steel structures according to claim 8, characterized in that, The process of dynamically adjusting the robot's stiffness and damping parameters based on the deviation between the expected and actual contact forces includes: Based on the digital twin, the stiffness and damping parameters of the robot end effector are dynamically adjusted by an adaptive impedance controller according to the deviation between the expected contact force and the actual contact force measured by the six-dimensional force sensor. When the actual contact force is greater than expected, the stiffness parameter is reduced and the damping parameter is increased to absorb excess energy. When the actual contact force is less than expected, the stiffness parameter is increased.

Images