A trajectory planning method for an intelligent service robot used in steel structure processing

By constructing a reduced-order thermo-coupled finite element model and fusing it with sparse point cloud Kalman filtering, combined with zero-space projection of redundant robotic arms, the problem of insufficient perception of global deformation of workpieces in steel structure processing was solved, realizing real-time prediction and safe replanning, and improving the continuity and reliability of processing.

CN123103694APending Publication Date: 2026-11-17HEILONGJIANG JINGZHI STEEL STRUCTURE CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611588449.3
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-10-14
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

Existing robot trajectory planning methods cannot perceive the global dynamic deformation of the workpiece in real time during steel structure processing, which makes the robotic arm prone to getting stuck in kinematically singular configurations or colliding and interfering with the workpiece body, causing sudden stops, workpiece scrapping, or even equipment damage.

Method used

By constructing a reduced-order thermo-coupled finite element deformation prediction proxy model and fusing it with Kalman filtering of sparse point cloud observation data, the global deformation of the workpiece is predicted in real time. Global replanning is then performed using the zero-space projection of the Jacobian matrix of the redundant robotic arm to avoid collisions and singular configurations.

Benefits of technology

It enables millisecond-level real-time prediction of global workpiece deformation, improving the continuity and reliability of processing, reducing protective emergency stops, and ensuring the safety of equipment and workpieces.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123103694A_ABST
    Figure CN123103694A_ABST
Patent Text Reader

Abstract

This invention discloses a trajectory planning method for an intelligent service robot used in steel structure processing. The invention relates to the field of intelligent robots and automated processing control technology. This method is based on a reduced-order thermo-coupled finite element deformation prediction proxy model. It predicts the global deformation field of the workpiece online under the coupling effect of thermal input and its own weight. The prediction results are then fused with sparse point cloud observation data collected from non-processing areas using Kalman filtering to obtain a fused estimated global dynamic deformation mesh. While maintaining process constraints such as the end effector's attitude envelope angle and end effector linear velocity, a collision risk potential field function and a singularity risk potential field function are constructed based on this deformation mesh. The pose of each joint of the robotic arm is adjusted in real time through the null-space projection of the redundant robotic arm's Jacobian matrix, avoiding collision interference or kinematic singularities caused by the workpiece's thermo-coupled deformation. This method can be applied to automated processing scenarios such as steel structure welding, cutting, and grinding, improving processing continuity and reliability.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of intelligent robots and automated processing control technology, specifically to a trajectory planning method for an intelligent service robot used in steel structure processing. Background Technology

[0002] In the manufacturing industry of large steel structures such as bridges, ships, and buildings, robotic automated processing is gradually replacing manual labor and is widely used in hot processing processes such as welding, cutting, and grinding. Typically, six-axis or seven-axis redundant industrial robots are used in conjunction with end effectors to complete continuous trajectory processing. Laser vision sensors and six-dimensional torque sensors are integrated at the end of the robotic arm to achieve local positioning and contact force feedback. Because the strong heat input during steel structure processing can cause large-scale, time-varying thermo-mechanical coupling deformation of the workpiece, the deformation can reach several centimeters or even tens of centimeters. Existing trajectory planning methods usually perform a one-time static point cloud scan of the workpiece before processing and register it with the CAD design model to generate an initial processing trajectory. Then, the local vision sensor or arc sensor integrated at the end of the robotic arm detects the local deviation of the weld seam and cutting line relative to the end effector. The local feedback control law such as proportional-integral-derivative is used to locally correct the end pose. Finally, the inverse kinematics solution is converted into joint angle commands to drive the movement of the robotic arm.

[0003] The above-mentioned solution only corrects local deviations at the end point and does not perceive and evaluate the global deformation state of the workpiece in real time. When the overall deformation of the workpiece accumulates to a certain extent, the end point pose correction required for local correction will cause the joint angle of the robot arm to deflect far beyond the initial planning range through inverse kinematic mapping. This will cause the robot arm to approach a kinematically singular configuration, or cause non-end point parts such as the elbow and wrist to collide and interfere with the deformed workpiece body, thereby triggering a protective emergency stop, resulting in processing interruption, workpiece scrapping, or even equipment damage. There is an urgent need for a trajectory planning method that can perceive the global dynamic deformation of the workpiece and collaboratively replan the global posture of the robot arm. Summary of the Invention

[0004] The technical problem this invention aims to solve is that during steel structure processing, the workpiece undergoes large-scale, time-varying thermo-mechanical coupling deformation due to strong heat input. Existing robot trajectory planning methods rely solely on local end-effector tracking for local correction, failing to perceive the global deformation state of the workpiece. This leads to the robotic arm easily falling into kinematically singular configurations or colliding and interfering with the deformed workpiece body while continuously deflecting along local feature points, resulting in sudden stops, workpiece scrapping, or even equipment damage. This invention aims to provide a trajectory planning method that can predict the global dynamic deformation of the workpiece in real time, integrate sparse visual observation data for correction, and replan the global posture of the robotic arm without compromising end-effector process constraints.

[0005] To address the aforementioned technical problems, this invention provides a trajectory planning method for an intelligent service robot used in steel structure processing, comprising:

[0006] Obtain the current process parameters, which include heat source power, processing speed, and the position of the heat source in the workpiece coordinate system;

[0007] The process parameters are substituted as thermal input excitation quantities into a pre-constructed reduced-order thermo-mechanical coupled finite element deformation prediction proxy model for online inference to obtain the prior prediction results of the global deformation field of the workpiece. The reduced-order thermo-mechanical coupled finite element deformation prediction proxy model compresses the full-node thermo-mechanical coupled finite element simulation results into low-dimensional modal coefficient dynamic equations based on the intrinsic orthogonal decomposition method.

[0008] Acquire sparse point cloud observation data of the non-processed area of ​​the workpiece by a three-dimensional laser profilometer installed at the end of the intelligent service robot or outside the processing fixture;

[0009] The prior prediction results and the sparse point cloud observation data are fused using Kalman filtering to obtain a fused estimated global dynamic deformation mesh.

[0010] Based on the coordinate position of the current processing point on the fusion estimated global dynamic deformation mesh and the preset end-effector process constraints, the expected velocity vector of the end effector of the intelligent service robot is calculated. The end-effector process constraints include the attitude envelope angle range and the end linear velocity.

[0011] The Jacobian matrix and its pseudo-inverse matrix are calculated based on the current joint angles of the intelligent service robot. The pseudo-inverse matrix is ​​multiplied by the desired velocity vector to obtain the first joint velocity component.

[0012] Based on the fusion estimation of the global dynamic deformation mesh, the distance between each link of the intelligent service robot body and the deformed workpiece surface is calculated, and a collision risk potential field function is constructed.

[0013] Based on the Jacobian matrix, the operability index is calculated, and the singularity risk potential field function is constructed.

[0014] The collision risk potential field function and the singularity risk potential field function are weighted and summed to obtain the composite risk potential field function.

[0015] The gradient of the composite risk potential field function with respect to the joint angle vector is calculated, and the zero-space self-motion joint velocity vector is obtained by calculating according to the preset step size coefficient.

[0016] Projecting the self-motion joint velocity vector in the null space onto the null space of the Jacobian matrix yields the second joint velocity component.

[0017] The first joint velocity component is added to the second joint velocity component to obtain the joint velocity command;

[0018] The joint speed command is output to the servo drive unit of the intelligent service robot to drive the movement of each joint of the intelligent service robot.

[0019] Compared with the prior art, the beneficial effects of the present invention by adopting the above technical solution are as follows:

[0020] Firstly, by constructing a reduced-order thermo-mechanical coupled finite element deformation prediction proxy model, and using process parameters as excitation input to perform online recursion on the low-dimensional modal coefficient dynamic equation, the shortcomings of the traditional one-time static scanning and CAD model matching method that cannot perceive dynamic deformation during processing are overcome, and millisecond-level real-time prediction of the global deformation field of the workpiece is achieved.

[0021] Secondly, by fusing the sparse point cloud observation data collected in the non-processing area with the prior prediction results using Kalman filtering, and adaptively reducing the dimension of the observation matrix when the observation point sampling fails, the defects of the end vision sensor being prone to failure and difficult to obtain global observation data under strong arc light and splashing environments are overcome, and low-cost, highly robust global deformation perception of the workpiece is achieved.

[0022] Third, by maintaining the desired velocity vector of the end effector unchanged, a collision risk potential field function and a singularity risk potential field function are constructed based on the fusion estimation of the global dynamic deformation mesh. The global joint pose of the robot arm is adjusted in real time by using the zero-space projection of the redundant robot arm Jacobian matrix. This overcomes the defects of blind joint deflection caused by local correction, which leads to singular configurations or body collisions. It achieves safe adjustment of the robot arm's global posture while ensuring the continuity of the end-effector processing, reduces protective emergency stops caused by singular configurations or collision interference, and improves the continuity and reliability of automated processing of steel structure robots. Attached Figure Description

[0023] Figure 1 This is a diagram showing the overall architecture of the intelligent service robot trajectory planning system of the present invention.

[0024] Figure 2 This is a block diagram illustrating the construction and online inference principle of the reduced-order thermo-coupled finite element deformation prediction proxy model of the present invention;

[0025] Figure 3 This is a flowchart of the Kalman filter fusion process of sparse point cloud observation data and workpiece dynamic deformation field in this invention;

[0026] Figure 4 This is a block diagram of the global replanning control principle based on zero-space redundancy in this invention;

[0027] Figure 5This is an overall flowchart of the trajectory planning method for intelligent service robots used in steel structure processing according to the present invention. Detailed Implementation

[0028] The following is in conjunction with the appendix Figure 1-5 The specific embodiments of the present invention will be further described below. It should be noted that the description of these embodiments is for the purpose of helping to understand the present invention, but does not constitute a limitation of the present invention. In addition, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0029] System Overall Architecture

[0030] Reference Figure 1 The overall architecture of the hardware and software system upon which this invention relies includes: a process parameter input unit, used to receive heat source power, processing speed, and the position of the heat source in the workpiece coordinate system, and output these as thermal input excitation quantities to a reduced-order thermo-coupled finite element deformation prediction proxy model; the reduced-order thermo-coupled finite element deformation prediction proxy model module, which receives the process parameters and outputs the prior prediction result of the global deformation field of the workpiece, i.e., the predicted deformation mesh; and a perception feedback link, consisting of a 3D laser profilometer installed at the end of the robotic arm or outside the processing fixture, used to collect sparse point clouds of the non-processed area of ​​the workpiece, and output the collection results to a Kalman filter fusion module; the input terminals of the Kalman filter fusion module are respectively connected to the reduced-order thermo-coupled finite element deformation prediction proxy model. The predicted output and the sparse point cloud observation output of the 3D laser profilometer are combined. The output is a corrected fusion estimation global dynamic deformable mesh, which is updated in real time and fed back to the global replanning module. The global replanning module takes as input the fusion estimation global dynamic deformable mesh, the current joint angles of the intelligent service robot, and the process constraint parameters of the end effector. The output is a joint speed command. The robot servo drive unit receives the joint speed command through the EtherCAT industrial bus and drives the seven-axis redundant robot body to move. At the same time, the six-dimensional torque sensor and the end effector local vision sensor feed back the contact force and local deviation information to the global replanning module in real time, forming a closed loop. The online calculation of the above modules is carried out by an industrial edge computing all-in-one machine.

[0031] Prior prediction of global dynamic deformation of workpiece

[0032] Reference Figure 2 This module is used to predict the global deformation field of a workpiece under the coupling effect of current thermal input and self-weight in real time within millisecond delay, replacing the traditional one-time static scanning and CAD model matching method. The construction process of this module is divided into an offline training stage and an online inference stage.

[0033] During the offline training phase, for multiple combinations of process parameters including different heat source power, processing speed and processing path, full-node thermo-coupling finite element simulation is used to generate a batch of workpiece full-time deformation field simulation sample datasets. The simulation sample datasets are subjected to intrinsic orthogonal decomposition to extract orthogonal modal basis vectors of a preset order. Based on the simulation sample datasets, the state transition matrix and input matrix of the reduced-order system are fitted with least squares to obtain the parameters of the modal coefficient dynamic equation.

[0034] This model approximates the deformation field of the workpiece as a linear combination of finite-order modal bases using the intrinsic orthogonal decomposition method:

[0035] ;

[0036] in, Indicates the position on the workpiece At any moment The deformation displacement field, Indicates the first The first-order orthogonal modal basis vectors, obtained from offline finite element simulation sample data through intrinsic orthogonal decomposition, reflect the spatial distribution of typical deformation modes of the workpiece. Indicates the first The time-varying coefficients corresponding to the first mode are low-dimensional state variables that need to be solved recursively online. This indicates that the number of modes retained is much smaller than the total number of degrees of freedom in the original finite element model.

[0037] The time-domain evolution of the modal coefficient vector is described by the following descending-order linear dynamic equation:

[0038] ;

[0039] in, Indicates by A state vector consisting of modal coefficients This represents the first derivative of the state vector with respect to time. This represents the current thermal input excitation vector, including heat source power, processing speed, and heat source location coordinates. This represents the state transition matrix of a reduced-order system. The input matrix of the reduced-order system is represented by both of them, which are determined by fitting offline simulation sample data and reflect the response law of the workpiece material thermodynamic properties and geometric structure to deformation.

[0040] In the online inference phase, the modal basis vector set and the reduced-order system matrix obtained in the offline phase are loaded, and the modal coefficient vector is initialized to the initial state before workpiece processing. During processing, the current process parameters are collected in real time at a fixed sampling period to form a thermal input excitation vector. The modal coefficient vector at the current moment and the thermal input excitation vector are substituted into the reduced-order dynamic equation, and the numerical integration method is used to further deduce to obtain the predicted modal coefficient value at the next moment. The deduce modal coefficient is linearly combined with the pre-loaded modal basis vector set to reconstruct the global deformation field prediction value of the workpiece at the next moment, and output to the Kalman filter fusion module in the form of a three-dimensional deformable mesh. The above acquisition, deduction and reconstruction process is executed cyclically until the processing task ends, thereby realizing continuous real-time prediction of the global deformation field of the workpiece.

[0041] Real-time fusion of sparse point clouds and dynamic deformation fields

[0042] Reference Figure 3 This module is used to correct the prediction results in real time based on the prior prediction results using sparse point cloud observation data of the non-processed area of ​​the workpiece, so as to solve the problem of global perception blind spots caused by the cumulative error of the prediction model and the failure of the end vision sensor under strong arc light and splash environment.

[0043] The mathematical principle of this fusion process adopts the Kalman filter recursive method, in the prediction stage:

[0044] ;

[0045] ;

[0046] in, Indicates the first Prior modal coefficient estimation at time t. Indicates the first Posterior modal coefficient estimation at time t. This represents the state transition matrix obtained by discretizing the reduced-order dynamic equations. Indicates the first The heat input excitation vector at time t. This represents the prior estimation error covariance matrix. Let represent the posterior estimation error covariance matrix of the previous time step. This represents the process noise covariance matrix, which reflects the uncertainties introduced by model simplification and simulation errors.

[0047] Update phase:

[0048] ;

[0049] ;

[0050] ;

[0051] in, Indicates the first The Kalman gain matrix at time 10:00. The observation matrix represents the space of modal coefficients, used to map the space of sparse point cloud observation points to a space of deformations. It is pre-calibrated and determined by the values ​​of the sampling points on the modal basis vectors. Indicates the first The sparse point cloud observation vectors actually collected at any given time. This represents the observation noise covariance matrix, reflecting the measurement error of the 3D laser profilometer. Indicates the fused and corrected number of... Time-posterior modal coefficient estimation, Represents the identity matrix.

[0052] In the specific implementation process, the observation matrix, process noise covariance, and observation noise covariance are determined in advance through offline calibration. In each sampling period, the predicted value of the prior modal coefficients and the prior covariance at the current moment are obtained first. The 3D laser profilometer performs sparse point cloud sampling on the preset feature points in the non-processing area of ​​the workpiece to obtain the measured observation vector. If there are observation points that fail to be sampled due to arc light interference and splash occlusion in the current period, the observation components corresponding to the observation points that fail to be sampled are marked as invalid, and the dimension of the observation matrix is ​​reduced accordingly. Only the remaining valid observation components are used in subsequent calculations. The Kalman gain matrix is ​​calculated based on the prior covariance, the observation matrix, and the observation noise covariance. The prior modal coefficients are corrected using the Kalman gain matrix to obtain the fused posterior modal coefficient estimate, and the posterior covariance matrix is ​​updated synchronously. The posterior modal coefficients are substituted into the modal basis vectors for reconstruction to obtain the corrected fused estimate of the global dynamic deformation mesh. This result is simultaneously fed back to the initial state of the next period's recursion and output to the global replanning module as the input for environmental perception. The above process is executed cyclically until the processing task is completed.

[0053] Zero-space redundancy-based process lossless global replanning

[0054] Reference Figure 4 This module is used to adjust the spatial pose of each joint of the robotic arm in real time by utilizing the zero-space self-motion capability of the redundant robotic arm, under the premise of ensuring the absolute stability of the process constraints of the end effector, and to avoid body collision or kinematic singularity problems caused by workpiece deformation. The process constraints mentioned here refer to the range of posture envelope angles and end linear velocity that any end tool among the welding torch, cutting torch and grinding head must maintain during the processing. This constraint remains unchanged during the trajectory replanning process.

[0055] The end-effector velocity and the velocities of each joint of the robotic arm satisfy the following kinematic relationship:

[0056] ;

[0057] in, This represents the desired velocity vector of the end effector, including the end position velocity and attitude angular velocity. This represents the vector of the robotic arm's current joint angle. The Jacobian matrix below, This represents the angular velocity vector of each joint.

[0058] For a robotic arm with redundant degrees of freedom, the general solution to the above equation is:

[0059] ;

[0060] in, Describes the pseudo-inverse of the Jacobian matrix. This refers to the first joint velocity component described in this invention, which corresponds to a set of joint velocity solutions that satisfy the end-effector velocity constraints. Let represent the null space projection matrix of the Jacobian matrix, which projects any joint velocity vector onto the self-motion subspace without changing the end-effector velocity. This represents the null-space self-motion joint velocity vector to be designed, used to adjust the global attitude of the robotic arm body without affecting the end-effector process. This null-space projection matrix is ​​then combined with... The product of the first and second joint velocity components is denoted as the second joint velocity component. Adding the first and second joint velocity components yields the joint velocity command. From the structure of this general solution, it can be seen that regardless of the value of the null space self-motion joint velocity vector, the joint velocity command obtained by superimposing it onto the first joint velocity component after null space projection always satisfies... This means that the actual speed of the end effector is always equal to the desired speed vector and is not affected by the zero-space self-motion component, thus ensuring that the continuity of the end-processing is not disrupted by the global replanning process.

[0061] The design of the null space self-motion vector is obtained by taking the negative gradient of the composite risk potential function:

[0062] ;

[0063] ;

[0064] in, The collision risk potential field function is constructed based on the distances between each link of the robotic arm and the workpiece surface, calculated in real time using a fusion-estimated global dynamic deformable mesh. The smaller this distance, the larger the function value. The singularity risk potential field function is constructed based on the reciprocal of the operability index of the Jacobian matrix. The closer the robotic arm is to the singular configuration, the larger the function value. , This represents the corresponding weighting coefficient, used to balance the relative importance of the two types of risks. Its value is preset to a specific value during implementation. This represents the composite risk potential field function. This represents the composite risk potential field function relative to the joint angle vector. gradient, This represents a preset positive step size coefficient used to control the adjustment rate of the zero-space self-motion.

[0065] In the specific implementation process, firstly, based on the actual coordinate position of the current processing point on the fused estimated global dynamic deformation mesh, and combined with the preset end-effector process constraints, the expected velocity vector of the end effector is calculated. This calculation process is consistent with the existing end-effector local tracking method. Then, based on the current joint angle vector, the Jacobian matrix and its pseudo-inverse matrix of the robotic arm are calculated to obtain the first joint velocity component. Next, based on the current fused estimated deformation mesh, the distances between each link of the robotic arm body, especially the elbow and wrist, and the deformed workpiece surface are calculated to construct a collision risk potential field function. At the same time, based on the current Jacobian matrix, the operability index is calculated to construct a singularity risk potential field function. The two are weighted and summed to obtain a composite risk potential field function. The composite risk potential field function is then used to calculate the velocity relative to the joint. The gradient of the angular vector is calculated, and the zero-space self-motion joint velocity vector is calculated according to the preset step size coefficient. The zero-space self-motion joint velocity vector is projected onto the Jacobian matrix zero space to obtain the second joint velocity component, which is added to the first joint velocity component to obtain the joint velocity command. It is determined whether the value of the current composite risk potential field function exceeds the preset safety threshold. If it does not exceed the threshold, the joint velocity command is directly output to the servo drive unit for execution. If it exceeds the preset safety threshold, the magnitude of the desired velocity vector is reduced by a preset ratio, and the above calculation process is repeated until the value of the composite risk potential field function falls back to within the safety threshold. The above process is repeated to continuously update the joint velocity command with a fixed control cycle until the current machining path segment is completed.

[0066] Overall Methodology and Flow

[0067] Reference Figure 5The overall process flow of this invention starts from the start of the processing task and sequentially executes the following steps: input of process parameters and prior prediction of global deformation field, acquisition of sparse point cloud and validity judgment, Kalman filter fusion, end-process constraint solution, risk potential field assessment and safety threshold judgment, zero-space self-motion solution and joint velocity synthesis, and joint velocity command output. It also determines whether the current processing path has been completed. If not, it returns to the process parameter input and prior prediction of global deformation field process to continue the loop. If it has been completed, it ends the current processing task. The above process flow completely and in a closed loop reflects the entire implementation process of this invention from process parameter input, deformation prediction, observation fusion, risk assessment to joint command output. Specific Implementation

[0068] Taking the automated robotic welding of a steel truss component as an example, a seven-axis redundant welding robot is used, with a welding torch as the end effector. The robotic arm integrates a laser positioning sensor and a six-dimensional torque sensor at its end. A three-dimensional laser profilometer is mounted externally on the machining fixture. Before machining, finite element simulation and intrinsic orthogonal decomposition modeling are performed for typical welding process parameter combinations of the component to obtain the modal basis vector set and the reduced-order system matrix. This data is then loaded onto an industrial edge computing all-in-one machine. During machining, the control system collects welding current, welding speed, and welding torch position in real time at a fixed sampling period. This data is input into the reduced-order dynamic equation for online recursion, obtaining the prior prediction result of the global deformation field of the workpiece. Simultaneously, the three-dimensional laser profilometer performs simultaneous analysis on the preset feature points in the non-machining areas of the component. Sparse point cloud acquisition is performed, and the results are fused with prior predictions using Kalman filtering to obtain a fused estimate of the global dynamic deformable mesh. The replanning module calculates the expected velocity vector and the first joint velocity component based on this deformable mesh and the current process constraints of the welding torch. At the same time, it constructs a collision risk potential field function and a singularity risk potential field function, solves the null space self-motion joint velocity vector, and projects it to obtain the second joint velocity component. The joint velocity command obtained by adding the two is sent to the servo drive unit through the EtherCAT bus to drive the movement of each joint of the robotic arm. Under the premise of ensuring that the welding torch attitude envelope angle and end effector linear velocity remain unchanged, the pose of the elbow and wrist of the robotic arm relative to the deformed component surface is adjusted in real time, avoiding singular configurations or body collisions caused by the accumulation of component thermal deformation.

[0069] Other implementation methods

[0070] In another embodiment of the present invention, the reduced-order thermo-mechanical coupled finite element deformation prediction proxy model can be constructed using a physical information neural network approach. That is, the heat conduction equation and the elastoplastic mechanics equation are used as physical constraints of the loss function to train a neural network model with process parameters as input and the deformation of key nodes of the workpiece as output. This can also achieve real-time prediction of the global deformation field of the workpiece, and can be combined with the aforementioned Kalman filter fusion process and zero-space replanning process to constitute an equivalent implementation of the present invention.

[0071] In another embodiment of the present invention, the global replanning process based on zero-space redundancy does not require the construction of a unified composite risk potential field function for gradient solving. Instead, it adopts a predictive collision detection method based on the envelopment of the robotic arm body links and the workpiece surface, which is based on the fusion estimation of the global dynamic deformation mesh. When it is predicted that a collision or near-singular configuration will occur within a predetermined number of consecutive control cycles, the joint velocity vector that satisfies the end-effector velocity constraint and avoids collisions and singular regions can be directly obtained by solving a constrained quadratic programming problem. This also enables global attitude adjustment while ensuring the continuity of the end-effector process, constituting an equivalent implementation of the present invention.

[0072] 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 possible changes and modifications without departing from the spirit and scope of the present invention. Therefore, any modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the content of the technical solution of the present invention shall fall within the protection scope defined by the claims of the present invention.

Claims

1. A trajectory planning method for an intelligent service robot used in steel structure processing, characterized in that, include: Obtain the current process parameters, which include heat source power, processing speed, and the position of the heat source in the workpiece coordinate system; The process parameters are substituted as thermal input excitation quantities into a pre-constructed reduced-order thermo-mechanical coupled finite element deformation prediction proxy model for online inference to obtain the prior prediction results of the global deformation field of the workpiece. The reduced-order thermo-mechanical coupled finite element deformation prediction proxy model compresses the full-node thermo-mechanical coupled finite element simulation results into low-dimensional modal coefficient dynamic equations based on the intrinsic orthogonal decomposition method. Acquire sparse point cloud observation data of the non-processed area of ​​the workpiece by a three-dimensional laser profilometer installed at the end of the intelligent service robot or outside the processing fixture; The prior prediction results and the sparse point cloud observation data are fused using Kalman filtering to obtain a fused estimated global dynamic deformation mesh. Based on the coordinate position of the current processing point on the fusion estimated global dynamic deformation mesh and the preset end-effector process constraints, the expected velocity vector of the end effector of the intelligent service robot is calculated. The end-effector process constraints include the attitude envelope angle range and the end linear velocity. The Jacobian matrix and its pseudo-inverse matrix are calculated based on the current joint angles of the intelligent service robot. The pseudo-inverse matrix is ​​multiplied by the desired velocity vector to obtain the first joint velocity component. Based on the fusion estimation of the global dynamic deformation mesh, the distance between each link of the intelligent service robot body and the deformed workpiece surface is calculated, a collision risk potential field function is constructed, an operability index is calculated based on the Jacobian matrix, a singularity risk potential field function is constructed, and the collision risk potential field function and the singularity risk potential field function are weighted and summed to obtain a composite risk potential field function. The gradient of the composite risk potential field function with respect to the joint angle vector is calculated, and the zero-space self-motion joint velocity vector is obtained by calculating according to the preset step size coefficient. Project the self-motion joint velocity vector in the null space onto the null space of the Jacobian matrix to obtain the second joint velocity component. Add the first joint velocity component and the second joint velocity component to obtain the joint velocity command. The joint speed command is output to the servo drive unit of the intelligent service robot to drive the movement of each joint of the intelligent service robot.

2. The method according to claim 1, characterized in that, The reduced-order thermo-coupling finite element deformation prediction proxy model is constructed in the following way: for multiple combinations of process parameters including different heat source power, processing speed and processing path, full-node thermo-coupling finite element simulation is performed to generate a simulation sample dataset of the workpiece's full-time deformation field. The simulation sample dataset is decomposed into intrinsic orthogonal components to extract orthogonal modal basis vectors of a preset order. Based on the simulation sample dataset, the reduced-order system state transition matrix and the reduced-order system input matrix are fitted to obtain the parameters of the modal coefficient dynamic equation.

3. The method according to claim 2, characterized in that, The step of substituting the process parameters into a pre-constructed reduced-order thermo-mechanical coupled finite element deformation prediction proxy model for online inference to obtain the prior prediction result of the global deformation field of the workpiece includes: substituting the modal coefficient vector and the thermal input excitation vector at the current moment into the modal coefficient dynamic equation, obtaining the predicted value of the modal coefficient at the next moment through numerical integration, and reconstructing the predicted value of the modal coefficient with the orthogonal modal basis vector through linear combination to obtain the prior prediction result.

4. The method according to claim 1, characterized in that, The step of fusing the prior prediction results with the sparse point cloud observation data using Kalman filtering to obtain a fused estimated global dynamic deformable mesh includes: using the modal coefficients corresponding to the prior prediction results as prior modal coefficient estimates; mapping the sparse point cloud observation data to modal coefficient observations through a pre-calibrated observation matrix; calculating the Kalman gain matrix based on the prior estimation error covariance matrix, the observation matrix, and the observation noise covariance matrix; correcting the prior modal coefficient estimates using the Kalman gain matrix to obtain posterior modal coefficient estimates; and reconstructing the posterior modal coefficient estimates by linear combination with the orthogonal modal basis vectors to obtain the fused estimated global dynamic deformable mesh.

5. The method according to claim 4, characterized in that, Also includes: If the three-dimensional laser profilometer fails to sample observation points due to arc interference and splash occlusion during the current sampling period, the observation components corresponding to the failed observation points are marked as invalid, and the dimension of the observation matrix is ​​reduced accordingly. The remaining valid observation components are then used to participate in the calculation of the Kalman gain matrix.

6. The method according to claim 1, characterized in that, During the calculation of the second joint velocity component and the addition of the first joint velocity component to the second joint velocity component to obtain the joint velocity command, the desired velocity vector remains unchanged.

7. The method according to claim 1, characterized in that, Also includes: Determine whether the value of the composite risk potential field function exceeds a preset safety threshold. If the value of the composite risk potential field function exceeds the preset safety threshold, reduce the magnitude of the desired velocity vector by a preset ratio, and recalculate the first joint velocity component, the composite risk potential field function, the zero-space self-motion joint velocity vector, the second joint velocity component, and the joint velocity command based on the reduced desired velocity vector, until the value of the composite risk potential field function does not exceed the preset safety threshold.

8. The method according to claim 1, characterized in that, The value of the collision risk potential field function increases as the distance decreases, and the singularity risk potential field function is constructed based on the reciprocal of the operability index, and its value increases as the intelligent service robot approaches a singular configuration.

9. The method according to claim 1, characterized in that, The intelligent service robot is a seven-axis redundant robot, and the end effector is any one of a welding torch, a cutting torch, and a grinding head.

10. A trajectory planning system for an intelligent service robot used in steel structure processing, characterized in that, include: The deformation prediction module is used to obtain the current process parameters, and substitute the process parameters as thermal input excitation quantities into the pre-constructed reduced-order thermo-mechanical coupled finite element deformation prediction proxy model for online inference to obtain the prior prediction results of the global deformation field of the workpiece. The reduced-order thermo-mechanical coupled finite element deformation prediction proxy model compresses the full-node thermo-mechanical coupled finite element simulation results into low-dimensional modal coefficient dynamic equations based on the intrinsic orthogonal decomposition method. A 3D laser profilometer is used to sample sparse point clouds in non-machined areas of a workpiece to obtain sparse point cloud observation data. The fusion module is used to fuse the prior prediction results with the sparse point cloud observation data through Kalman filtering to obtain a fused estimated global dynamic deformation mesh. The replanning module is used to calculate the expected velocity vector of the end effector based on the coordinate position of the current processing point on the fused estimated global dynamic deformation mesh and the preset end process constraints. It calculates the Jacobian matrix and its pseudo-inverse matrix based on the current joint angles of the intelligent service robot to obtain the first joint velocity component. It constructs a composite risk potential field function based on the fused estimated global dynamic deformation mesh and the Jacobian matrix and obtains the gradient with respect to the joint angle vector to obtain the null space self-motion joint velocity vector. It projects the null space self-motion joint velocity vector onto the null space of the Jacobian matrix to obtain the second joint velocity component. It adds the first joint velocity component and the second joint velocity component to obtain the joint velocity command. A servo drive unit is used to receive the joint speed commands and drive the movement of each joint of the intelligent service robot.