Intelligent multi-head welding synchronous control system

Abstract

The invention relates to the technical field of control adjustment, in particular to an intelligent multi-head welding synchronous control system which comprises a data processing module, a modular cooperative control module, a load sensing module, a data transmission module, a self-adaptive pose execution module and a fault-tolerant module. The data processing module extracts three-dimensional geometrical characteristics of a workpiece through laser scanning, PID control parameters are migrated across working conditions in combination with a dynamic time warping algorithm, the inverse kinematics process is optimized, and the self-adaptive pose execution module drives phase synchronization of multiple welding heads based on a global reference trajectory and predicts and compensates thermal deformation errors through a thermal expansion coefficient. And the fault-tolerant module adopts a wavelet packet decomposition identification fault mode, and updates a historical parameter library in combination with an elastic weight algorithm. According to the method, three-dimensional space pose error accumulation caused by mechanical load fluctuation and signal asynchronization is inhibited through cooperation of multiple modules, and the welding process uniformity and the system dynamic stability are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of control and regulation, and particularly to an intelligent multi-head welding synchronization control system. Background Art

[0002] The multi-head welding synchronization control system is based on industrial automation standards, and coordinates the working states of multiple welding devices by integrating a sensor network and a central control unit. The system uses real-time data acquisition technology to monitor the displacement, temperature and motion trajectory of each welding head, and dynamically adjusts welding parameters by means of a closed-loop feedback mechanism. When a position deviation or heat input deviation occurs in a certain welding unit, the control algorithm will reallocate the drive signals of the actuators, so that all welding heads maintain a predetermined phase difference in the space coordinate system, thereby eliminating the material deformation caused by the superposition of multiple heat sources. Through the two core strategies of timing synchronization and energy balance, this architecture realizes the uniformity of the welding process while avoiding mechanical interference, and meets the high-efficiency forming requirements of complex components.

[0003] In the dynamic disturbance scenario of the multi-head welding synchronization control system, it is challenging to maintain the phase consistency of the multi-axis motion trajectories, mainly due to the asymmetric distribution of mechanical load differences between welding units and sensor signal transmission delays: when the system performs complex space curve welding, the inertial torque fluctuations of each welding head will cause the response speed of the actuators to be discretized, and the fusion time difference of multi-source heterogeneous data in the closed-loop feedback channel exceeds the control cycle threshold, resulting in the inability of the cooperative algorithm based on the automatic control architecture to fully compensate for the cumulative pose deviation of the execution end in the three-dimensional space, ultimately affecting the spatio-temporal coupling accuracy of the multi-source heat energy field. Summary of the Invention

[0004] Aiming at the deficiencies of the prior art, the present invention provides an intelligent multi-head welding synchronization control system, which solves the problem of cumulative three-dimensional space pose synchronization error caused by dynamic fluctuations of mechanical loads and asynchronous transmission of multi-source signals in the multi-head welding system.

[0005] To solve the above technical problems, the specific technical solutions of the present invention are as follows: The present invention provides an intelligent multi-head welding synchronization control system, including: A data processing module, configured to extract three-dimensional geometric features and material thermodynamics parameters of the workpiece through laser scanning and spectral analysis, and perform spatio-temporal alignment and coordinate fusion on multi-sensor data to generate a feature matrix containing weld path planning information; A modular collaborative control module, receiving the feature matrix output by the data processing module, generating hierarchical task instructions based on the path planning information in the feature matrix, extracting load dynamic features through edge computing, and combining the load dynamic features with the inverse kinematic solution mapping to generate a smooth trajectory instruction in the joint space; The load perception module receives the trajectory instruction sent by the modular collaborative control module, fuses real-time force, current, and vibration signals to generate a multi-modal load feature vector, dynamically loads control parameters from the historical model library through cross-condition migration matching, and feeds back the control parameters to the inverse kinematic mapping process of the modular collaborative control module; The data transmission module is respectively connected to the data processing module, the modular collaborative control module, and the load perception module, prioritizes the sensor data of the data processing module and performs hardware-level verification, and synchronously transmits the verified sensor data to the modular collaborative control module and the load perception module; The adaptive pose execution module receives the trajectory instruction generated by the modular collaborative control module and the sensor data synchronized by the data transmission module, generates a phase synchronization instruction according to the global reference trajectory, and adjusts the trajectory coordinates based on the thermal deformation compensation amount fed back by the load perception module to drive the collaborative movement of the welding head; The fault tolerance module monitors the pose feedback signal of the adaptive pose execution module and the fault characteristics of the load perception module, dynamically reconstructs the control parameters and updates the control parameters in the historical model library.

[0006] Furthermore, for the intelligent multi-head welding synchronization control system of the present invention, the data processing module includes: The laser scanning unit is configured to segment the weld seam area based on the region growing algorithm and extract the curvature continuity feature, and transmit the segmented three-dimensional geometric feature to the spectral analysis unit; The spectral analysis unit is configured to receive the three-dimensional geometric feature output by the laser scanning unit, generate a set of material thermodynamics parameters by matching the emission spectrum with the thermodynamics parameter map in the material database; The spatio-temporal alignment unit is configured to receive the set of material thermodynamics parameters output by the spectral analysis unit, align the visual positioning data and the force sensor signal in terms of time stamp in the global coordinate system, and output a feature matrix containing three-dimensional path coordinates and thermodynamics attributes to the modular collaborative control module.

[0007] Furthermore, for the intelligent multi-head welding synchronization control system of the present invention, the modular collaborative control module includes: The task planning unit is configured to receive the feature matrix output by the data processing module, generate a welding path including obstacle avoidance constraints based on the rapidly-exploring random tree algorithm, and call the heat conduction model in combination with the material thermodynamics parameters in the feature matrix to generate a power distribution instruction for multiple welding heads; The edge computing unit is configured to receive the multi-source association matrix output by the feature fusion unit, extract the frequency domain feature components of the vibration signal through a convolutional neural network, and synchronously analyze the harmonic distortion rate of the servo current signal; The motion mapping unit is configured to receive the power distribution instruction of the task planning unit and the frequency-domain characteristic components of the edge computing unit, introduce a harmonic compensation coefficient when converting the Cartesian space trajectory instruction into a joint angle sequence, and generate a servo drive signal with vibration suppression through cubic spline interpolation.

[0008] Furthermore, in the intelligent multi-head welding synchronization control system of the present invention, the load perception module includes: The feature fusion unit is configured to receive the force sensor signal, motor current signal, and accelerometer signal verified by the data transmission module, perform dimensionality reduction processing on the multi-source time-domain signals through principal component analysis, and generate a load feature vector containing the main energy components; The parameter matching unit is configured to receive the load feature vector output by the feature fusion unit, use the dynamic time warping algorithm to compare the current welding trajectory with the working condition characteristic curves stored in the historical model library, and when the similarity exceeds the threshold, call the corresponding PID control parameter set and inject it into the interpolation algorithm of the motion mapping unit.

[0009] Furthermore, in the intelligent multi-head welding synchronization control system of the present invention, the data transmission module includes: The time-sensitive network unit is configured to receive the sensor data stream processed by the data verification unit, and according to the real-time level of the control system, use the credit shaping algorithm to assign the highest transmission priority to the pose feedback data; The data verification unit is configured to perform cyclic redundancy check on the original sensor data at the FPGA hardware layer, perform bit error correction processing on the sampling points with abnormal verification using Hamming codes, and generate clean data packets with time stamps and distribute them to the load perception module and the adaptive pose execution module.

[0010] Furthermore, in the intelligent multi-head welding synchronization control system of the present invention, the adaptive pose execution module includes: The phase synchronization unit is configured to receive the trajectory instruction generated by the modular collaborative control module, generate a global reference trajectory with a time stamp through the virtual master controller, and drive each slave welding head to perform collaborative motion according to a preset phase difference; The thermal compensation unit is configured to receive the multi-modal load feature vector feedback by the load perception module, calculate the predicted thermal deformation value of the welding head based on the material thermal expansion coefficient look-up table, and generate a trajectory coordinate compensation amount and inject it into the reference trajectory generation process of the phase synchronization unit.

[0011] Furthermore, in the intelligent multi-head welding synchronization control system of the present invention, the fault tolerance module includes: The fault diagnosis unit is configured to receive the multi-modal load feature vector output by the load perception module, extract the high-frequency transient component of the current signal by using wavelet packet decomposition, and identify two fault modes of mechanical jamming and sensor failure through a pre-trained decision tree classifier; The model update unit is configured to receive the fault feature vector of the fault diagnosis unit, dynamically update the control parameter set of the historical model library in the double buffer storage area, and use the elastic weight consolidation algorithm to constrain the gradient change direction during parameter update to maintain the stability of the existing control strategy.

[0012] Advantages of the present invention; Through the spatio-temporal alignment and multi-source data fusion of the data processing module, the present invention generates a feature matrix containing thermodynamic attributes, providing multi-dimensional parameter inputs for path planning; the modular collaborative control module combines the rapidly-exploring random tree algorithm and the heat conduction model to dynamically allocate welding power, and extracts frequency domain features through edge computing to suppress the trajectory deviation caused by load fluctuations; the load perception module realizes the cross-condition PID parameter migration based on the principal component analysis and the dynamic time warping algorithm, and dynamically optimizes the inverse kinematic mapping process; the data transmission module uses hardware-level verification and time-sensitive network to ensure the synchronous transmission of multi-source signals and eliminate asynchronous errors; the adaptive pose execution module corrects the trajectory coordinates through phase synchronization and thermal deformation compensation, and combines the fault diagnosis of the fault tolerance module and the elastic weight algorithm to maintain the stability of the control strategy. Finally, the three-dimensional space pose synchronization accuracy of multiple welding heads is improved under mechanical load fluctuations and thermal disturbances, enhancing the welding process uniformity and system reliability under complex working conditions. Description of the drawings

[0013] In order to more clearly illustrate the technical solutions of the present invention, the drawings required for the embodiments will be briefly introduced below. Obviously, for those of ordinary skill in the art, other drawings can be obtained according to the drawings without creative efforts.

[0014] Figure 1 It is a system architecture diagram of an intelligent multi-head welding synchronization control system provided by an embodiment of the present invention. Detailed implementation manners

[0015] To make the objectives, technical solutions, and advantages of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below in conjunction with the specific embodiments and corresponding drawings of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention. The technical solutions provided by each embodiment of the present invention are described in detail below in conjunction with the drawings. To better understand the objectives of the present invention, the present invention is further described in detail below.

[0016] Please refer to Figure 1 , the present invention provides an intelligent multi-head welding synchronization control system, including: A data processing module, which is used to extract the three-dimensional geometric features and material thermodynamics parameters of the workpiece through laser scanning and spectral analysis, perform spatio-temporal alignment and coordinate fusion on multi-sensor data, and generate a feature matrix containing weld path planning information; A modular collaborative control module, which receives the feature matrix output by the data processing module, generates hierarchical task instructions based on the path planning information in the feature matrix, extracts the load dynamic features through edge computing, combines the load dynamic features with the inverse kinematic solution mapping, and generates a joint space smooth trajectory instruction; A load sensing module, which receives the trajectory instruction sent by the modular collaborative control module, fuses real-time force, current and vibration signals to generate a multi-modal load feature vector, dynamically loads control parameters from the historical model library through cross-condition migration matching, and feeds back the control parameters to the inverse kinematic solution mapping process of the modular collaborative control module; A data transmission module, which is respectively connected to the data processing module, the modular collaborative control module and the load sensing module, divides the priority of the sensor data of the data processing module and performs hardware-level verification, and synchronously transmits the verified sensor data to the modular collaborative control module and the load sensing module; An adaptive pose execution module, which receives the trajectory instruction generated by the modular collaborative control module and the sensor data synchronized by the data transmission module, generates a phase synchronization instruction according to the global reference trajectory, and adjusts the trajectory coordinates based on the thermal deformation compensation amount fed back by the load sensing module to drive the welding heads to move collaboratively; A fault tolerance module, which monitors the pose feedback signal of the adaptive pose execution module and the fault features of the load sensing module, dynamically reconstructs the control parameters and updates the control parameters in the historical model library.

[0017] In the intelligent multi - head welding synchronization control system provided by the present invention, the data processing module uses a laser scanning unit to collect three - dimensional point clouds of the workpiece surface, identifies the weld boundary based on the region - growing algorithm and extracts the curvature continuity features, and transmits the segmented geometric features to the spectral analysis unit for material property analysis. The spectral analysis unit generates a dataset containing key parameters such as thermal conductivity and specific heat capacity by matching the emission spectrum of the workpiece with the thermodynamic parameter maps in the pre - stored material database, providing input for subsequent heat conduction modeling. The spatio - temporal alignment unit receives the thermodynamic parameter set output by the spectral analysis unit, combines the spatial coordinates and force sensor signals collected by the vision positioning system, eliminates the timing deviation during multi - source data acquisition through timestamp alignment, and generates a feature matrix containing three - dimensional path coordinates and thermodynamic properties in the global coordinate system. This matrix serves as the path planning basis for the modular collaborative control module.

[0018] After receiving the feature matrix, the task planning unit of the modular collaborative control module uses the rapidly - exploring random tree algorithm to generate a welding path topology map that satisfies the obstacle - avoidance constraints, calls the finite - element heat conduction model in combination with the material thermodynamic parameters, and calculates the power distribution strategy for multiple welding heads to balance the heat input. The edge - computing unit synchronously receives the multi - source correlation matrix of the load - sensing module, extracts the frequency - domain energy distribution features of the vibration signal through a convolutional neural network, and analyzes the harmonic distortion rate of the servo - motor current signal to identify load fluctuations. The motion mapping unit combines the power distribution instruction output by the task planning unit with the frequency - domain features of the edge - computing unit, introduces a harmonic compensation coefficient when converting the Cartesian - space trajectory into a joint - angle sequence, and generates a servo - drive signal with vibration - suppression function through cubic spline interpolation to ensure the smooth movement of the end - effector of the robotic arm.

[0019] The feature fusion unit of the load - sensing module receives the signals from the force sensor, current sensor, and accelerometer after being verified by the data transmission module, eliminates the redundant information of the multi - source time - domain signals through principal - component analysis, and extracts the core feature vector representing the mechanical load state. The parameter matching unit performs dynamic time warping matching between the current load feature vector and the working - condition feature curves in the historical model library. When the trajectory similarity exceeds the set threshold, it automatically calls the corresponding working - condition PID control parameter set and injects it into the interpolation algorithm of the motion mapping unit to achieve cross - working - condition control parameter migration. The migrated parameters adjust the inverse kinematic mapping process in real - time through the feedback channel, forming a closed - loop parameter optimization mechanism.

[0020] The data verification unit of the data transmission module performs cyclic redundancy check on the original sensor data at the FPGA hardware layer, corrects bit errors for the sampling points with abnormal verification using Hamming code, and generates a clean data packet with a millisecond-level timestamp. The time-sensitive network unit assigns the highest transmission priority to the pose feedback data according to the real-time requirements of the control system using the credit shaping algorithm to ensure low-latency transmission of key data within the control cycle. The verified multi-modal data packet is synchronously distributed to the load perception module and the adaptive pose execution module to provide a consistent data benchmark for real-time control.

[0021] After receiving the trajectory instruction from the modular collaborative control module, the phase synchronization unit of the adaptive pose execution module generates a global reference trajectory with a synchronous clock through the virtual master controller, and drives each slave welding head to execute collaborative motion according to the preset phase difference. The thermal compensation unit continuously receives the multi-modal feature vector feedback from the load perception module, predicts the thermal deformation amount of the welding area based on the material thermal expansion coefficient look-up table, generates a trajectory coordinate compensation amount and injects it into the reference trajectory generation link of the phase synchronization unit to achieve online compensation for thermal deformation. The compensated trajectory instruction drives the servo mechanism to adjust the spatial pose of the welding head to maintain the spatio-temporal coupling accuracy of the multi-heat source energy field.

[0022] The fault diagnosis unit of the fault tolerance module continuously monitors the multi-modal feature vector output by the load perception module, extracts the high-frequency transient component of the current signal using wavelet packet decomposition, and identifies two fault modes of mechanical jamming and sensor failure through a pre-trained decision tree classifier. After detecting the fault characteristics, the model update unit uses a double-buffer storage mechanism to update the historical model library online, constrains the gradient change direction during the reconstruction of control parameters through the elastic weight consolidation algorithm, and maintains the stability of the original control strategy while absorbing new working condition data. The updated parameter set is synchronized to each control unit through the data transmission module to form the self-healing ability of the system.

[0023] Specifically, for the intelligent multi-head welding synchronization control system described in the present invention, the data processing module includes: The laser scanning unit is configured to segment the weld seam area based on the region growing algorithm and extract the curvature continuity feature, and transmit the segmented three-dimensional geometric feature to the spectral analysis unit; The spectral analysis unit is configured to receive the three-dimensional geometric feature output by the laser scanning unit, and generate a set of material thermodynamic parameters by matching the emission spectrum with the thermodynamic parameter map in the material database; The spatio-temporal alignment unit is configured to receive the set of material thermodynamic parameters output by the spectral analysis unit, align the visual positioning data and the force sensor signal in terms of timestamp in the global coordinate system, and output a feature matrix containing three-dimensional path coordinates and thermodynamic attributes to the modular collaborative control module.

[0024] In the data processing module, the laser scanning unit performs region segmentation on the three-dimensional point cloud of the workpiece surface based on the region growing algorithm, identifies the geometric features of the weld boundary through curvature continuity detection, and the segmented three-dimensional geometric model contains the spatial topology information of the weld path. This unit transmits the feature data set containing the curvature gradient to the spectral analysis unit, providing a structural basis for material property analysis. After receiving the three-dimensional geometric model, the spectral analysis unit uses the spectral matching algorithm to compare the peak values of the collected emission spectra of metal elements with the thermodynamic parameter maps in the material database, extracts the material thermal conductivity, specific heat capacity, and phase change temperature parameters, and generates a parameter data set containing thermodynamic properties. After receiving the thermodynamic parameter set output by the spectral analysis, the spatio-temporal alignment unit aligns the three-dimensional coordinate data of the vision positioning system and the contact force signal of the force sensor at the millisecond level through the timestamp synchronization mechanism, and generates a composite feature matrix containing spatial coordinates, contact force distribution, and thermodynamic properties under the global coordinate system. This feature matrix marks the heat input sensitive area of the weld path through the coordinate mapping relationship, providing multi-dimensional parameter input for the path optimization of the modular collaborative control module.

[0025] The region growing algorithm of the laser scanning unit starts iterative expansion from the point cloud seed points, judges the weld area boundary according to the curvature change rate threshold of adjacent points, and the generated geometric segmentation result contains the curvature gradient distribution feature. During the segmentation process, the curvature tolerance threshold of region growing is dynamically adjusted to adapt to the geometric mutation characteristics of different welding joints. The spectral analysis unit uses multi-channel filtering technology to eliminate ambient light interference during the spectral matching stage, calculates the material thermodynamic parameters through the characteristic wavelength intensity ratio, and preferentially compares the isomorphic alloy maps in the material database during the matching process. The spatio-temporal alignment unit establishes a coordinate transformation model between the vision positioning system and the force sensor, uses quaternion interpolation to compensate for the sampling frequency difference of multi-sensors, and generates a fusion data stream with timestamp synchronization, where the vision data provides the absolute coordinates of the weld path, the force sensor data reflects the contact pressure distribution, and the thermodynamic parameters mark the heat sensitivity coefficient of the path points.

[0026] After completing the material parameter extraction, the spectral analysis unit correlates and marks the thermal conductivity parameter with the curvature feature of the geometric model, generating a three-dimensional path point set with thermal property labels. The spatio-temporal alignment unit unifies the multi-source data to the base coordinate system of the welding robot through the coordinate transformation matrix, uses Kalman filtering to eliminate the coordinate drift caused by sensor noise, and the output feature matrix contains the spatial coordinates, normal vector direction, contact force value, and thermal conductivity of each path point. This matrix is transmitted to the task planning unit of the modular collaborative control module through the data bus, providing multi-physical field coupling parameter input for the dynamic obstacle avoidance planning of the welding path and the heat source power distribution, forming a full-process data processing link from geometric recognition to control parameter generation.

[0027] Specifically, for the intelligent multi-head welding synchronization control system of the present invention, the modular collaborative control module includes: The task planning unit is configured to receive the feature matrix output by the data processing module, generate a welding path including obstacle avoidance constraints based on the rapidly-exploring random tree algorithm, and call the heat conduction model in combination with the material thermodynamics parameters in the feature matrix to generate a power distribution instruction for multiple welding heads; The edge computing unit is configured to receive the multi-source correlation matrix output by the feature fusion unit, extract the frequency-domain feature components of the vibration signal through a convolutional neural network, and synchronously analyze the harmonic distortion rate of the servo current signal; The motion mapping unit is configured to receive the power distribution instruction of the task planning unit and the frequency-domain feature components of the edge computing unit, introduce a harmonic compensation coefficient when converting the Cartesian space trajectory instruction into a joint angle sequence, and generate a servo drive signal with vibration suppression through cubic spline interpolation.

[0028] After receiving the feature matrix output by the data processing module, the task planning unit of the modular collaborative control module generates a welding path topology map that satisfies the minimum safety distance in three-dimensional space based on the rapidly-exploring random tree algorithm. During the path planning process, the material thermal conductivity parameters marked in the feature matrix are called in real time, and the heat accumulation trend of each welding area is calculated in combination with the finite element heat diffusion model to generate a dynamic power distribution instruction for multiple welding heads. This instruction includes the power gradient change curve of each welding head at the path points and is used to control the balance of heat input distribution. After receiving the multi-source correlation matrix output by the feature fusion unit of the load perception module, the edge computing unit extracts the features of the time-frequency spectrum of the vibration acceleration signal through a convolutional neural network, captures the feature components in the frequency-domain energy concentration interval, and synchronously analyzes the harmonic distortion rate of the three-phase current signal of the servo motor to identify the load fluctuation mode of the mechanical transmission system.

[0029] After receiving the power distribution instruction output by the task planning unit and the frequency-domain feature components extracted by the edge computing unit, the motion mapping unit dynamically adjusts the compensation coefficient according to the harmonic distortion rate data during the conversion process from the Cartesian space trajectory to the joint space angle sequence to suppress mechanical resonance in a specific frequency band. The cubic spline interpolation algorithm is used in the conversion process to smooth the discrete path points and generate a servo drive signal sequence with vibration suppression function. This signal sequence controls the angular displacement of the servo motor through pulse width modulation to achieve continuous and smooth movement of the end effector of the welding head in the joint space. The power distribution instruction acts on the pulse width modulation module of the welding power supply synchronously, so that the heat input intensity of multiple welding heads forms a dynamic matching relationship with the mechanical movement speed.

[0030] The edge computing unit adopts a sliding window mechanism to process real-time data streams during the frequency-domain feature analysis stage. The window width is matched with the natural frequency range of the mechanical system to ensure the timeliness of feature extraction and the frequency-domain resolution. The convolutional neural network compresses the high-frequency noise components of the vibration signal through multiple pooling operations, focusing on the low-frequency features that reflect the mechanical load state. The motion mapping unit introduces joint angular acceleration constraint conditions during the interpolation process to prevent trajectory jitter caused by rapid acceleration and deceleration. At the same time, the harmonic compensation coefficient is written into the feedforward control channel of the servo drive to form a composite control strategy for vibration suppression. The collaborative optimization of the power distribution instruction and the motion parameters enables the dynamic coupling of the heat source energy distribution and the mechanical motion trajectory in the time and space dimensions.

[0031] Specifically, for the intelligent multi-head welding synchronization control system described in the present invention, the load sensing module includes: The feature fusion unit is configured to receive the force sensor signal, the motor current signal, and the accelerometer signal verified by the data transmission module, and perform dimensionality reduction processing on the multi-source time-domain signals through principal component analysis to generate a load feature vector containing the main energy components. The parameter matching unit is configured to receive the load feature vector output by the feature fusion unit, and use the dynamic time warping algorithm to compare the current welding trajectory with the working condition feature curves stored in the historical model library. When the similarity exceeds the threshold, the corresponding PID control parameter set is called and injected into the interpolation algorithm of the motion mapping unit.

[0032] After receiving the force sensor, motor current, and accelerometer signals verified by the data transmission module, the feature fusion unit of the load sensing module calculates the covariance matrix of the multi-source time-domain signals through principal component analysis, extracts the main energy components of each signal channel, eliminates the redundant features caused by sensor noise, and generates a core feature vector representing the mechanical load state. This vector contains the linear combination weight coefficients of the force-current-vibration signals, reflecting the dynamic load distribution of the mechanical structure during the welding process. The feature fusion unit uses a sliding window mechanism to frame the real-time signal stream, and the window length is matched with the welding process cycle to ensure the timeliness of feature extraction.

[0033] After receiving the load feature vector output by the feature fusion unit, the parameter matching unit calculates the morphological similarity between the current welding trajectory feature curve and the working condition templates stored in the historical model library through the dynamic time warping algorithm. The algorithm elastically matches the sequence of trajectory points on the time axis to eliminate the timing deviation caused by the welding speed fluctuation. When the similarity score exceeds the preset threshold, the PID control parameter set corresponding to the working condition is retrieved from the historical model library. The successfully matched parameter set is injected into the cubic spline interpolation algorithm of the motion mapping unit through the data bus to adjust the gain coefficient of the servo drive signal. During the parameter migration process, a weighted fusion strategy is adopted to mix the historical parameters and the current real-time parameters according to the similarity ratio, avoiding the step change of the control parameters.

[0034] The dynamic time warping algorithm establishes a cost matrix in the trajectory comparison stage, solves the minimum cumulative distance path through dynamic programming, and introduces a welding path curvature weight factor during the matching process to enhance the comparison accuracy of spatial geometric features. The historical model library adopts a hierarchical storage structure, and stores the working condition feature curves and the corresponding control parameter sets classified by material type, welding speed, and joint form. The parameter injection link is synchronized with the interpolation period of the motion mapping unit, and the PID parameters are written into the feedforward control channel of the servo driver during the trajectory point discretization stage, forming a closed-loop parameter optimization mechanism based on load state perception. The dynamic mapping relationship between the feature vector and the control parameters enables the system to adapt to the mechanical load fluctuations under different welding working conditions.

[0035] Specifically, for the intelligent multi-head welding synchronization control system described in the present invention, the data transmission module includes: The time-sensitive network unit is configured to receive the sensor data stream processed by the data verification unit, and according to the real-time level of the control system, uses the credit shaping algorithm to assign the highest transmission priority to the pose feedback data; The data verification unit is configured to perform cyclic redundancy verification on the original sensor data at the FPGA hardware layer, perform bit error correction processing on the sampling points with abnormal verification using Hamming codes, and generate clean data packets with time stamps and distribute them to the load perception module and the adaptive pose execution module.

[0036] The data verification unit of the data transmission module performs cyclic redundancy check on the original sensor data at the FPGA hardware layer, calculates the check code through a preset polynomial to verify data integrity, and uses Hamming code to correct single-bit errors for the sampled points with verification anomalies, generating clean data packets with millisecond-level timestamps. The timestamp generation mechanism is synchronized with the system global clock to eliminate the time reference deviation of multi-sensor data. The error-corrected data packets are transmitted to the time-sensitive network unit through a high-speed bus, providing a standardized format input for subsequent data transmission. The time-sensitive network unit divides data priorities according to the real-time requirements of the control system, and uses a credit shaping algorithm to allocate a dedicated transmission channel for pose feedback data, and preferentially schedules high-priority data frames within a fixed time window to achieve low-latency transmission of critical control data.

[0037] The data verification unit adopts a parallel computing architecture to accelerate the generation of check codes during the cyclic redundancy check stage, processes 32-bit data blocks per clock cycle, and the verification anomaly triggers the enable signal of the Hamming code error correction circuit. The error correction logic locates and repairs the flipped errors of data bits through exclusive OR operations, and attaches a timestamp mark accurate to the microsecond level to the repaired data packet. The time-sensitive network unit maintains a priority mapping table, maps pose feedback data, vibration signals, and current signals to three transmission levels of emergency, high, and medium respectively, and the credit shaping algorithm allocates bandwidth quotas according to the levels, and the emergency-level data enjoys preemptive transmission rights. The processed data stream is synchronously transmitted to the load perception module and the adaptive pose execution module through a dual-channel distribution mechanism. Among them, the pose data is directly injected into the phase synchronization unit, and the vibration and current signals enter the feature fusion unit to form a data synchronization reference for closed-loop control.

[0038] The time-sensitive network unit dynamically adjusts the credit value allocation strategy within the scheduling cycle. The credit pool capacity of the emergency-level data is set to 3 times that of the high-level data to ensure the transmission of pose feedback data within the control cycle. The mirror transmission technology is adopted during the data distribution process to achieve simultaneous replication and split transmission of data packets at the physical layer, eliminating the timing error caused by serial transmission. The timestamp information of the clean data packet is clock-aligned with the global reference trajectory generation module of the adaptive pose execution module, so that the sensor data and the control instructions of the actuator are strictly synchronized in the time dimension, supporting the coordinated motion control of multiple welding heads.

[0039] Specifically, for the intelligent multi-head welding synchronization control system described in the present invention, the adaptive pose execution module includes: The phase synchronization unit is configured to receive the trajectory instruction generated by the modular collaborative control module, generate a global reference trajectory with a timestamp through the virtual master controller, and drive each slave welding head to perform collaborative motion according to a preset phase difference; The thermal compensation unit is configured to receive the multi-modal load feature vector fed back by the load perception module, calculate the predicted thermal deformation value of the welding head based on the material thermal expansion coefficient look-up table, and generate a trajectory coordinate compensation amount to be injected into the reference trajectory generation process of the phase synchronization unit.

[0040] After receiving the trajectory instruction generated by the modular collaborative control module, the phase synchronization unit of the adaptive pose execution module analyzes the spatio-temporal coordinate sequence of the path points through the virtual master controller, and generates a global reference trajectory with millisecond-level timestamps in combination with the system global clock. This reference trajectory is synchronously transmitted to the motion controllers of each slave welding head through the EtherCAT bus. The slave controller performs time offset processing on the reference trajectory according to the preset phase difference parameter, and drives the multi-welding heads to maintain a synchronous motion rhythm in the spatial dimension. The phase difference parameter is dynamically adjusted according to the welding path curvature. An equal phase difference strategy is adopted in the straight section, and the phase difference is reduced according to the curvature radius in the inflection point area to prevent trajectory interference.

[0041] The thermal compensation unit receives in real time the multi-modal load feature vector fed back by the load perception module, analyzes the welding area temperature gradient data contained therein, and matches the linear expansion coefficient of the current workpiece based on the material thermal expansion coefficient look-up table. The predicted thermal deformation value of the end of the welding head is calculated through a thermal-mechanical coupling model, and a three-dimensional space coordinate compensation amount is generated. When the compensation amount is injected into the reference trajectory generation link of the phase synchronization unit, a feed-forward compensation mechanism is adopted to convert the predicted deformation amount into a trajectory point offset instruction, and the thermal deformation error is corrected in advance during the path planning stage. The compensated reference trajectory updates the target pose of the welding head through the coordinate transformation matrix, so that the actual welding path dynamically adapts to the thermal deformation characteristics of the material.

[0042] When generating the global reference trajectory, the phase synchronization unit establishes a master-slave station clock synchronization mechanism. The master controller periodically sends synchronization frames to calibrate the slave station clock, and the timestamp accuracy is controlled at the microsecond level to meet the requirements of high-speed welding. The material thermal expansion coefficient look-up table of the thermal compensation unit is stored classified by material type, and the parameter set corresponding to the material number of the current welding workpiece is preferentially called during the matching process. An environmental temperature correction factor is introduced during the compensation amount calculation to eliminate the influence of workshop temperature drift on the thermal deformation prediction. The corrected trajectory coordinates are converted into joint driving amounts through kinematic forward solutions, and the servo motors are driven to perform collaborative motion with thermal compensation to maintain the spatial pose consistency of the multi-welding heads under thermal disturbances.

[0043] Specifically, for the intelligent multi-head welding synchronization control system of the present invention, the fault tolerance module includes: The fault diagnosis unit is configured to receive the multi-modal load feature vector output by the load perception module, extract the high-frequency transient components of the current signal by using wavelet packet decomposition, and identify two fault modes of mechanical jamming and sensor failure through a pre-trained decision tree classifier; The model update unit is configured to receive the fault feature vector of the fault diagnosis unit, dynamically update the control parameter set of the historical model library in the double-buffer storage area, and use the elastic weight consolidation algorithm to constrain the gradient change direction during parameter update to maintain the stability of the existing control strategy.

[0044] After receiving the trajectory instruction generated by the modular collaborative control module, the phase synchronization unit of the adaptive pose execution module analyzes the spatio-temporal coordinate sequence of the path points through the virtual master controller, and generates a global reference trajectory with millisecond-level timestamps in combination with the system global clock. This reference trajectory is synchronously transmitted to the motion controllers of each slave station welding head through the EtherCAT bus. The slave station controller performs time offset processing on the reference trajectory according to the preset phase difference parameter, and drives the multi-welding heads to maintain synchronous motion beats in the spatial dimension. The phase difference parameter is dynamically adjusted according to the curvature of the welding path. The equal phase difference strategy is adopted in the straight line segment, and the phase difference is reduced according to the curvature radius in the inflection point area to prevent trajectory interference.

[0045] The thermal compensation unit receives the multi-modal load feature vector fed back by the load perception module in real time, analyzes the welding area temperature gradient data contained therein, and matches the linear expansion coefficient of the current workpiece based on the material thermal expansion coefficient look-up table. The predicted value of the thermal deformation at the end of the welding head is calculated through a thermal-mechanical coupling model, and a three-dimensional space coordinate compensation amount is generated. When the compensation amount is injected into the reference trajectory generation link of the phase synchronization unit, a feed-forward compensation mechanism is used to convert the predicted deformation amount into a trajectory point offset instruction, and the thermal deformation error is corrected in advance during the path planning stage. The compensated reference trajectory updates the target pose of the welding head through the coordinate transformation matrix, so that the actual welding path dynamically adapts to the thermal deformation characteristics of the material.

[0046] The phase synchronization unit establishes a master-slave station clock synchronization mechanism when generating the global reference trajectory. The master station controller periodically sends synchronization frames to calibrate the slave station clock, and the timestamp accuracy is controlled at the microsecond level to meet the requirements of high-speed welding. The thermal expansion coefficient look-up table of the thermal compensation unit is stored by material type classification, and the parameter set corresponding to the material number of the current welding workpiece is preferentially called during the matching process. An environmental temperature correction factor is introduced during the compensation amount calculation to eliminate the influence of workshop temperature drift on the thermal deformation prediction. The corrected trajectory coordinates are converted into joint driving amounts through kinematic forward solution, and the servo motors are driven to perform collaborative motion with thermal compensation to maintain the spatial pose consistency of the multi-welding heads under thermal disturbances.

[0047] The present invention solves the problem of cumulative error in three-dimensional spatial pose synchronization through a multi-module collaborative control architecture. First, the data processing module extracts the three-dimensional geometric features and material thermodynamics parameters of the workpiece through laser scanning and spectral analysis. The spatio-temporal alignment unit synchronizes the timestamps of the visual positioning data and the force sensor signals to generate a feature matrix containing thermodynamic attributes. This matrix provides multi-physical field coupling parameters for the modular collaborative control module. The task planning unit generates an obstacle avoidance path based on the rapidly-exploring random tree algorithm, and dynamically allocates the power of multiple welding heads in combination with the heat conduction model. The edge computing unit analyzes the frequency domain characteristics of the vibration signal and the current harmonic distortion rate through a convolutional neural network, and extracts the load fluctuation characteristics in real time. The motion mapping unit introduces a harmonic compensation coefficient to generate a smooth trajectory command, suppressing the trajectory deviation caused by the dynamic fluctuation of the mechanical load.

[0048] Secondly, the feature fusion unit of the load perception module performs principal component analysis and dimensionality reduction on the force, current, and vibration signals to generate a core feature vector representing the load state. The parameter matching unit compares the current working condition with the historical model library through the dynamic time warping algorithm, and dynamically loads the optimal PID parameters to the motion mapping unit to achieve cross-working condition control parameter migration. The data transmission module implements cyclic redundancy check and Hamming code error correction at the FPGA hardware layer. The time-sensitive network unit uses the credit shaping algorithm to assign the highest transmission priority to the pose data, eliminating the timing error of asynchronous transmission of multi-source signals and providing a synchronous data reference for closed-loop control.

[0049] Finally, the phase synchronization unit of the adaptive pose execution module drives multiple welding heads to move collaboratively with a phase difference based on the global reference trajectory. The thermal compensation unit predicts the thermal deformation amount according to the material thermal expansion coefficient and generates a trajectory coordinate compensation amount to correct the deformation error. The fault tolerance module identifies mechanical jamming and sensor failure faults through wavelet packet decomposition and decision tree classifier, and updates the parameters of the historical model library using the elastic weight consolidation algorithm, maintaining the stability of the control strategy while absorbing new working condition data, forming a dynamic error suppression and self-healing ability, and ensuring the spatial pose synchronization accuracy of multiple welding heads under complex working conditions.

Claims

1. An intelligent multi-head welding synchronization control system, characterized in that, It includes: A data processing module, which is used to extract the three-dimensional geometric features and material thermodynamics parameters of the workpiece through laser scanning and spectral analysis, perform spatio-temporal alignment and coordinate fusion on multi-sensor data, and generate a feature matrix containing weld path planning information; A modular collaborative control module, which receives the feature matrix output by the data processing module, generates hierarchical task instructions based on the path planning information in the feature matrix, extracts the load dynamic features through edge computing, and combines the load dynamic features with the inverse kinematic solution mapping to generate a smooth trajectory instruction in the joint space; A load perception module, which receives the trajectory instruction sent by the modular collaborative control module, fuses real-time force, current and vibration signals to generate a multi-modal load feature vector, dynamically loads control parameters from the historical model library through cross-condition migration matching, and feeds back the control parameters to the inverse kinematic solution mapping process of the modular collaborative control module; A data transmission module, which is respectively connected to the data processing module, the modular collaborative control module and the load perception module, divides the priority of the sensor data of the data processing module and performs hardware-level verification, and synchronously transmits the verified sensor data to the modular collaborative control module and the load perception module; An adaptive pose execution module, which receives the trajectory instruction generated by the modular collaborative control module and the sensor data synchronized by the data transmission module, generates a phase synchronization instruction according to the global reference trajectory, and adjusts the trajectory coordinates based on the thermal deformation compensation amount fed back by the load perception module to drive the coordinated movement of the welding head; A fault tolerance module, which monitors the pose feedback signal of the adaptive pose execution module and the fault characteristics of the load perception module, dynamically reconstructs the control parameters and updates the control parameters in the historical model library.

2. The intelligent multi-head welding synchronization control system according to claim 1, wherein The data processing module includes: The laser scanning unit is configured to segment the weld area based on the region growing algorithm and extract the curvature continuity feature, and transmit the segmented three-dimensional geometric feature to the spectral analysis unit; The spectral analysis unit is configured to receive the three-dimensional geometric feature output by the laser scanning unit, and generate a set of material thermodynamics parameters by matching the emission spectrum with the thermodynamics parameter map in the material database; The spatio-temporal alignment unit is configured to receive the set of material thermodynamics parameters output by the spectral analysis unit, align the time stamps of the visual positioning data and the force sensor signal in the global coordinate system, and output a feature matrix containing three-dimensional path coordinates and thermodynamic properties to the modular collaborative control module.

3. The intelligent multi-head welding synchronization control system according to claim 2, characterized in that, The modular collaborative control module includes: The task planning unit is configured to receive the feature matrix output by the data processing module, generate a welding path containing obstacle avoidance constraints based on the rapid exploration random tree algorithm, and call the heat conduction model in combination with the material thermodynamics parameters in the feature matrix to generate a power distribution instruction for multiple welding heads; The edge computing unit is configured to receive the multi-source correlation matrix output by the feature fusion unit, extract the frequency domain feature components of the vibration signal through a convolutional neural network, and synchronously analyze the harmonic distortion rate of the servo current signal; The motion mapping unit is configured to receive the power distribution instruction from the task planning unit and the frequency-domain feature components of the edge computing unit, introduce a harmonic compensation coefficient when converting the Cartesian space trajectory instruction into a joint angle sequence, and generate a servo drive signal with vibration suppression through cubic spline interpolation.

4. The intelligent multi-head welding synchronization control system according to claim 3, wherein The load sensing module includes: The feature fusion unit is configured to receive the force sensor signal, motor current signal, and accelerometer signal verified by the data transmission module, perform dimensionality reduction processing on the multi-source time-domain signals through principal component analysis, and generate a load feature vector containing the main energy components. The parameter matching unit is configured to receive the load feature vector output by the feature fusion unit, use the dynamic time warping algorithm to compare the current welding trajectory with the working condition feature curves stored in the historical model library, and when the similarity exceeds the threshold, call the corresponding PID control parameter set and inject it into the interpolation algorithm of the motion mapping unit.

5. The intelligent multi-head welding synchronization control system according to claim 4, wherein, The data transmission module includes: The time-sensitive network unit is configured to receive the sensor data stream processed by the data verification unit, and according to the real-time level of the control system, use the credit shaping algorithm to assign the highest transmission priority to the pose feedback data. The data verification unit is configured to perform cyclic redundancy check on the original sensor data at the FPGA hardware layer, perform bit error correction processing on the sampling points with abnormal checks using Hamming codes, and generate clean data packets with timestamps and distribute them to the load sensing module and the adaptive pose execution module.

6. The intelligent multi-head welding synchronization control system according to claim 5, characterized in that, The adaptive pose execution module includes: The phase synchronization unit is configured to receive the trajectory instruction generated by the modular collaborative control module, generate a global reference trajectory with a timestamp through the virtual master controller, and drive each slave welding head to perform collaborative motion with a preset phase difference. The thermal compensation unit is configured to receive the multi-modal load feature vector fed back by the load sensing module, calculate the predicted thermal deformation value of the welding head based on the material thermal expansion coefficient lookup table, and generate a trajectory coordinate compensation amount and inject it into the reference trajectory generation process of the phase synchronization unit.

7. The intelligent multi-head welding synchronization control system according to claim 6, wherein The fault tolerance module includes: The fault diagnosis unit is configured to receive the multi-modal load feature vector output by the load sensing module, extract the high-frequency transient components of the current signal using wavelet packet decomposition, and identify two fault modes: mechanical jamming and sensor failure through a pre-trained decision tree classifier. The model update unit is configured to receive the fault feature vector of the fault diagnosis unit, dynamically update the control parameter set of the historical model library in the double-buffer storage area, and use the elastic weight consolidation algorithm to constrain the gradient change direction during parameter update.

Images