Robot control multi-axis collaborative data real-time synchronous optimization method and system

Abstract

The invention provides a real-time synchronous optimization method and system for multi-axis collaborative data controlled by a robot, and relates to the technical field of robotics.The method comprises the steps that a collaborative motion control network is established based on a unified global clock source, poses, moments and process data of all axes of the robot are synchronously collected, and global timestamps are added; obtaining a data set with a timestamp; performing time alignment processing on the data set with the timestamp to generate a multi-axis synchronous original data stream; and based on a preset extended kinematics model containing an external axis, carrying out fusion calculation on the spatial pose data in the multi-axis synchronous original data stream, and carrying out real-time filtering on the moment and process data to obtain a collaborative state data set containing an expected state. Through the full-link cooperative control method from data acquisition, processing and transmission to optimization, the motion precision, self-adaptability and reliability of the multi-axis robot are improved.

Description

Technical Field

[0001] This invention relates to the field of robotics, and in particular to a method and system for real-time synchronization optimization of multi-axis collaborative data in robot control. Background Technology

[0002] As high-end manufacturing moves towards intelligence and flexibility, industrial robot applications have evolved from single-machine operations to closely collaborative multi-axis (jointed axes, external track axes, positioner axes, etc.) operation modes. In such tasks, real-time synchronization and collaborative control of the pose, torque, and process data of each axis are crucial for ensuring operational accuracy and process quality. Current mainstream solutions adopt a centralized control architecture, where the main controller processes data from each axis and issues commands. However, with increasing system complexity, this architecture has shortcomings at the data processing level, hindering further improvements in collaborative performance.

[0003] First, in the data analysis and state estimation stages, existing methods process data for each axis in isolation, lacking a global spatial perspective. For example, load filtering parameters are fixed, and multi-axis pose and load information are not correlated in real time. This makes it impossible to dynamically identify uneven load distribution and abrupt changes in the workspace caused by geometric and physical characteristics, resulting in insufficient adaptive capability of control parameters and potential vibration or tracking errors at critical process points. Second, in the control command transmission stage, the optimized command set relies on a single communication link for issuance. In complex industrial environments, link interruptions or data packet loss can disrupt multi-axis collaborative synchronization. Furthermore, existing systems lack a highly reliable mechanism to ensure the continuity of command flow, which may fail to meet the stringent requirements of uninterrupted collaboration in continuous processes. Finally, in the feedback data utilization stage, the system's utilization of actual operating data is mostly limited to monitoring alarms and local corrections. It fails to provide closed-loop feedback of deep features such as synchronization deviations and response delays to the data fusion and parameter optimization process of the next cycle. This means the system lacks the ability to perform feedforward self-calibration based on historical performance, which may prevent continuous optimization of collaborative performance. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to provide a method and system for real-time synchronization optimization of multi-axis collaborative data in robot control. Through a full-link collaborative control method from data acquisition, processing, transmission to optimization, the motion accuracy, adaptability and reliability of multi-axis robots are improved.

[0005] To solve the above-mentioned technical problems, the technical solution of the present invention is as follows: A first aspect includes a method for real-time synchronization optimization of multi-axis collaborative data in robot control, the method comprising: A cooperative motion control network is established based on a unified global clock source. The pose, torque and process data of each axis of the robot are collected synchronously and a global timestamp is added to obtain a time-stamped data set. The time-stamped data set is then time-aligned to generate a multi-axis synchronous raw data stream. Based on a pre-defined extended kinematic model including external axes, spatial pose data in multi-axis synchronous raw data streams are fused and calculated, and torque and process data are filtered in real time to obtain a cooperative state dataset containing the desired state. Based on the cooperative state dataset, the control parameters of each axis are solved synchronously and the load and pose information of key points are extracted to obtain a preliminary control parameter set; the load distribution in the spatial domain is analyzed by combining the load and pose information of key points to obtain the load uniformity characteristics; the load adaptive feedforward gain is obtained based on the load uniformity characteristics; the preliminary control parameter set and the load adaptive feedforward gain are fused to obtain an adaptive multi-axis cooperative motion command set with sequence identification. The adaptive multi-axis cooperative motion command set is sent to each axis driver through the primary and backup redundant communication links. The link status is monitored and the link is switched when there is an abnormality. The interruption resume is executed according to the sequence identifier to realize the reliable sending of the command set. Based on the reliable instruction set, actual operation feedback data is collected, compared with the expected state, and then the synchronization deviation and response delay are calculated and monitored and alarmed. The synchronization deviation and response delay are used as the input for the next cycle to drive a new round of collaborative processing.

[0006] Secondly, a robot control multi-axis collaborative data real-time synchronization optimization system includes: The acquisition module is used to establish a cooperative motion control network based on a unified global clock source, synchronously acquire the pose, torque and process data of each axis of the robot and attach a global timestamp to obtain a time-stamped data set; the time-stamped data set is then processed by time alignment to generate a multi-axis synchronous raw data stream; The data processing module is used to perform fusion calculations on the spatial pose data in the multi-axis synchronous raw data stream based on a preset extended kinematic model that includes external axes, and to perform real-time filtering on torque and process data to obtain a cooperative state dataset containing the desired state. The control module is used to synchronously calculate the control parameters of each axis and extract the load and pose information of key points based on the cooperative state dataset to obtain a preliminary control parameter set; combine the load and pose information of key points to perform spatial domain load distribution analysis to obtain load uniformity characteristics; obtain the load adaptive feedforward gain based on the load uniformity characteristics; and fuse the preliminary control parameter set and the load adaptive feedforward gain to obtain an adaptive multi-axis cooperative motion command set with sequence identifiers. The switching module is used to send the adaptive multi-axis cooperative motion command set to each axis driver through the primary and backup redundant communication links, monitor the link status and switch the link when abnormal, and perform breakpoint resume transmission according to the sequence identifier to realize the reliable sending of the command set. The closed-loop verification module is used to collect actual operation feedback data based on the reliably issued instruction set, compare it with the expected state, calculate the synchronization deviation and response delay, and monitor and alarm. The synchronization deviation and response delay are used as the input for the next cycle to drive a new round of collaborative processing.

[0007] The above-described solution of the present invention has at least the following beneficial effects: A unified global clock source provides a time reference for data acquisition, and global timestamps ensure that the timing of data from each axis is unique and correlated. Time alignment processing eliminates the acquisition delay difference between different axes, and the generated multi-axis synchronous raw data stream has high timing consistency, laying a reliable foundation for subsequent multi-axis data fusion. Extending the kinematic model to incorporate external axis parameters broadens the coverage of pose data fusion and improves the completeness of spatial pose calculation. Real-time filtering specifically suppresses high-frequency noise in torque and process data, and the resulting stable parameters, together with accurate pose, constitute a collaborative state dataset, providing high-quality data input for control parameter calculation. The synchronous calculation mechanism ensures the timing matching of control parameters for each axis, and the extraction of key point load and pose data focuses on core motion characteristics. Spatial domain load distribution analysis quantifies load uniformity characteristics, enabling the load adaptive feedforward gain to match the actual load state. Preliminary control parameters and... The system optimizes instruction adaptability through integration, assigning each instruction a unique traceability attribute through sequence identifiers, thus enhancing the controllability of instruction execution. Dual transmission guarantees are established through redundant primary and backup communication links, preventing data interruption due to single link failures. Link status monitoring enables real-time anomaly detection and rapid switching, and combined with the breakpoint resumption mechanism of sequence identifiers, it can accurately locate the interruption point and restore transmission, ensuring no packet loss or duplication during instruction set distribution, guaranteeing the reliability and continuity of instruction transmission. The comparison between actual operational feedback data and expected states is precisely correlated based on timestamps, ensuring the targeted nature of deviation calculations. Quantitative calculations of synchronization deviations and response delays provide an intuitive basis for operational status assessment, and the monitoring and alarm mechanism enables timely warnings of anomalies. Deviation data serves as input for the next cycle, forming a closed-loop feedback, allowing the collaborative processing flow to be dynamically adjusted based on historical operational data, continuously optimizing multi-axis collaborative performance. Attached Figure Description

[0008] Figure 1 This is a flowchart illustrating a method for real-time synchronization optimization of multi-axis collaborative data in robot control, provided by an embodiment of the present invention.

[0009] Figure 2 This is a schematic diagram of a robot control multi-axis collaborative data real-time synchronization optimization system provided by an embodiment of the present invention. Detailed Implementation

[0010] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

[0011] like Figure 1 As shown in the figure, an embodiment of the present invention proposes a method for real-time synchronization optimization of multi-axis collaborative data in robot control, the method comprising the following steps: Step 100: Establish a cooperative motion control network based on a unified global clock source, synchronously collect the pose, torque and process data of each axis of the robot and add a global timestamp to obtain a time-stamped data set; generate a multi-axis synchronous raw data stream by time alignment processing of the time-stamped data set. Step 200: Based on the preset extended kinematic model including the external axis, the spatial pose data in the multi-axis synchronous raw data stream is fused and calculated, and the torque and process data are filtered in real time to obtain a cooperative state dataset containing the desired state. Step 300: Based on the cooperative state dataset, synchronously calculate the control parameters of each axis and extract the load and pose information of key points to obtain a preliminary control parameter set; combine the load and pose information of key points to perform spatial domain load distribution analysis to obtain load uniformity characteristics; obtain the load adaptive feedforward gain based on the load uniformity characteristics; fuse the preliminary control parameter set and the load adaptive feedforward gain to obtain an adaptive multi-axis cooperative motion command set with sequence identifiers. Step 400: The adaptive multi-axis cooperative motion command set is sent to each axis driver through the primary and backup redundant communication links. The link status is monitored and the link is switched when there is an abnormality. The interruption resume is executed according to the sequence identifier to realize the reliable sending of the command set. Step 500: Collect actual operation feedback data based on the reliably issued instruction set, compare it with the expected state, calculate the synchronization deviation and response delay, and monitor and alarm. Use the synchronization deviation and response delay as the input for the next cycle to drive a new round of collaborative processing.

[0012] In this embodiment of the invention, a unified global clock source provides a consistent time reference for collaborative control, timestamps clearly define the data acquisition time, and time alignment processing eliminates time deviations in the data of each axis, enabling the original data stream of multi-axis synchronization to have precise time correlation, providing synchronization assurance for subsequent multi-axis collaborative calculations; the extended kinematic model incorporates external axis information to improve the integrity of spatial pose data fusion, and real-time filtering effectively reduces noise interference in torque and process data, making the collaborative state dataset more consistent with actual operating conditions and providing a reliable basis for control decisions; synchronous calculation ensures the correlation of control parameters of each axis, key point information extraction focuses on core operating characteristics, and the spatial domain negative... Load distribution analysis clearly reveals the load change pattern, and the load adaptive feedforward gain makes the control parameters more matched to the load state. The fused instruction set can adapt to the dynamic operation requirements of multi-axis. The primary and backup redundant communication links reduce the risk of instruction transmission interruption, and the link status monitoring and abnormal switching ensure transmission continuity. The breakpoint resume transmission supported by sequence identification avoids duplicate or missing instructions, improving the reliability and efficiency of instruction set issuance. The comparison between actual operation feedback data and expected state can accurately capture synchronization deviation and response delay. Real-time monitoring and alarms enable timely warning of abnormal situations. Deviation and delay data serve as input for the next cycle, forming a closed loop in the collaborative processing flow and continuously optimizing the multi-axis synchronization effect.

[0013] In a preferred embodiment of the present invention, step 100 above involves establishing a cooperative motion control network based on a unified global clock source, synchronously collecting the pose, torque, and process data of each axis of the robot and attaching a global timestamp to obtain a timestamped data set; and then performing time alignment processing on the timestamped data set to generate a multi-axis synchronous raw data stream, including: Step 101: Construct a unified global clock source and establish a cooperative motion control network based on the global clock source. Configure a synchronous clock reference and a unique axis identifier for the sensor data channels corresponding to each robot axis in the cooperative motion control network. Specifically, this includes: selecting a temperature-controlled crystal oscillator clock module with an accuracy of not less than 10ns as the core device of the unified global clock source; integrating this clock module onto the core control board of the robot's main controller; and amplifying and filtering the power of the original clock signal output by the clock module through a clock signal amplification circuit to form a stable global synchronous clock signal. Construct a cooperative motion control network based on the EtherCAT industrial communication protocol, configuring the robot's main controller as the master node of the network, and configuring the sensors and drivers corresponding to each robot axis as slave nodes. Physical communication links are established between each slave node via shielded twisted-pair cables, with a link transmission rate set to 1000Mbps to ensure data transmission bandwidth. Each slave node in the cooperative motion control network is assigned a unique network address, and a synchronization clock reference and a unique axis identifier are configured for the sensor data channels corresponding to each axis of the robot. The synchronization clock reference is calibrated by the master node periodically sending clock synchronization frames to the slave nodes, with a calibration period of 1ms. The axis identifier uses 8-bit binary encoding, with the first 3 bits representing the axis type and the last 5 bits representing the axis number under that type. In the axis type encoding, 001 represents a joint axis, 010 represents an external track walking axis, and 011 represents a servo positioner axis, ensuring that the axis identifier of each sensor data channel is uniquely identifiable in the entire cooperative motion control network.

[0014] Step 102: Through the cooperative motion control network, real-time pose, torque, and welding process data of each joint axis of the robot, the external track walking axis, and the servo positioner axis are synchronously collected to obtain a multi-axis raw data set. Specifically, after the cooperative motion control network is initialized, the master node sends a data acquisition start command to each slave node. This command includes the acquisition period, data type identifier, and trigger signal parameters. The acquisition period is set to 0.1ms according to process requirements, and the data type identifiers correspond to pose data, torque data, and welding process data, respectively. After receiving the start command, each slave node starts the corresponding data acquisition operation according to the command parameters. Specifically, each joint axis of the robot is connected to the joint electric field via a servo positioner. The absolute encoder at the end of the machine collects real-time pose data, including joint angles, angular velocities, and angular accelerations. The angle measurement accuracy is controlled within ±0.001°, and the angular velocity measurement range is 0 to 500° / s. The external track axis collects real-time pose data via a linear grating ruler, including travel position, linear velocity, and linear acceleration. The position measurement accuracy is controlled within ±0.01mm, and the linear velocity measurement range is 0 to 1m / s. The servo positioner axis collects real-time pose data via an incremental encoder, including displacement angles and displacement angular velocities. The angle measurement accuracy is controlled within ±0.005°. Torque data for each axis is collected by a torque sensor integrated into the motor output shaft, with a torque measurement range of 0 to 500N. m, with a measurement accuracy of ±0.5N. m; Welding process data is acquired through the signal acquisition module of the welding power source, including welding current, welding voltage, wire feed speed and shielding gas flow rate. The current measurement range is 50 to 300A, the voltage measurement range is 10 to 30V, the wire feed speed measurement range is 0.5 to 10m / min, and the shielding gas flow rate measurement range is 10 to 30L / min. After each slave node synchronously completes the acquisition of the corresponding data according to the preset acquisition cycle, the pose data, torque data and welding process data of the same axis are initially packaged to form a single-axis data unit. All single-axis data units are summarized to form a multi-axis raw data set.

[0015] Step 103: Based on the global clock source, append a global timestamp to the multi-axis raw data set to obtain a timestamped data set. Specifically, this includes: the master node sending the current clock signal generated by the global clock source to each slave node in real time through the cooperative motion control network. This clock signal uses a Unix timestamp format, accurate to the microsecond level. Each slave node has a data buffer unit internally. After a single-axis data unit completes acquisition and initial packaging, it immediately extracts the current global timestamp from the received clock signal and appends this global timestamp directly to the header of the single-axis data unit. The timestamp and the single-axis data unit are separated by frame division. Frame separators are used for differentiation. The frame separator is set to a specific 16-bit binary sequence to avoid confusion with the data content. After each slave node completes the timestamp appending, the single-axis data unit with timestamp is subjected to CRC-32 cyclic redundancy check. If the check passes, it is marked as a valid data unit. If the check fails, the corresponding data is re-acquired and the timestamp is appended again until a valid data unit is generated. After all the valid data units of the slave nodes are summarized, a timestamped data set containing each axis data and the corresponding global timestamp is formed. Each data unit contains five parts: axis identifier, global timestamp, data type, data length, and data body.

[0016] Step 104 involves uploading the timestamped data set to the data receiving area of ​​the robot control cabinet in real time via the cooperative motion control network. Specifically, each slave node uploads its timestamped data set to the main controller via the cooperative motion control network. The upload uses a priority-based polling mechanism, with the slave node corresponding to welding process data having the highest priority, followed by the slave node corresponding to joint axis data, and the slave node corresponding to external track walking axis and servo positioner axis data having the lowest priority. The main node receives the data uploaded by each slave node sequentially according to this priority order. A data fragmentation mechanism is used during data transmission. When the data length of a single axis data unit exceeds 1024 bytes... When data is transmitted in bytes, it is automatically divided into multiple data fragments. Each fragment header contains a fragment sequence number, total number of fragments, and data unit identifier to ensure complete data transmission. The data receiving area of ​​the robot control cabinet is set as a circular buffer with a buffer capacity of 10MB. This receiving area is divided into partitions according to axis identifiers. Each partition corresponds to a slave node of an axis. Independent read and write pointers are used between partitions to avoid conflicts during data writing and reading. The data receiving area receives data fragments uploaded by each slave node in real time. After all fragments of the same data unit have been received, the fragments are automatically reassembled. After reassembly, the partition is marked as pending processing and the receiving status identifier of the partition is updated.

[0017] Step 105: Receive the timestamped data sets from the data receiving area. Perform multi-source timing alignment processing based on the global timestamps carried by each data set to obtain a time-synchronized data sequence. Specifically, this includes: the timing processing module built into the robot's main controller scans the data units marked as pending processing in the data receiving area in real time, extracting the global timestamp and axis identifier carried in the header of each data unit; using the global timestamp of each data unit as a reference, determine the time reference range for this timing alignment. This range is the interval between the minimum and maximum timestamps of all currently scanned data units. Divide this time reference range into multiple time scales according to the acquisition cycle, with each time scale corresponding to a acquisition moment; for each time scale... The system retrieves data from all data units corresponding to axis identifiers to determine if a global timestamp exists for that time scale. If no data unit corresponds to an axis identifier at that time scale, linear interpolation is used to interpolate the pose, torque, and welding process data of the axis at two adjacent acquisition times to obtain the completed data for that time scale. If multiple data units exist for an axis identifier at that time scale, the data unit with the smallest deviation between its timestamp and the time scale is selected as the valid data, and the remaining data units with larger deviations are discarded. After completing the data retrieval, completion, and filtering for all time scales, a time-synchronized data sequence containing all axis data for each time scale is formed, and the timestamps of all axis data in this data sequence remain consistent.

[0018] Step 106 involves integrating the time-synchronized data sequence along the time axis to obtain a spatiotemporally unified multi-axis synchronized raw data stream. Specifically, the main controller's data stream integration module receives the time-synchronized data sequence and first verifies its validity. This verification includes checking whether the numerical range of each axis's data is within a preset threshold and whether the data format conforms to specifications. If a data point exceeds the preset threshold or has an incorrect format, it is marked as abnormal, and the corresponding axis identifier and time scale are recorded. Simultaneously, the corresponding data from the previous valid time scale is used for temporary replacement. After verification, the time-synchronized data sequence is sorted in ascending order of timestamps, using the time scale as an index, ensuring a continuous distribution of the data sequence along the time axis. Based on this sorting, the pose data, torque data, and welding process data of different axes at the same time scale are associated and bound with their corresponding axis identifiers, forming multi-axis data groups based on time scales. Each multi-axis data group contains a timestamp, each axis identifier, and complete data for the corresponding axis. Finally, all multi-axis data groups are sequentially concatenated along the time axis to generate a spatiotemporally unified multi-axis synchronized raw data stream.

[0019] In this embodiment of the invention, a unified global clock source provides a consistent time reference for data from each axis, and a unique axis identifier enables precise differentiation of data from each axis. The configuration of the synchronized clock reference and identifier lays the foundation for subsequent correlation processing of multi-axis data. Synchronous acquisition covers robot joint axes, external track walking axes, and servo positioner axes, ensuring the integrity of the multi-axis raw data set. Simultaneously, pose, torque, and welding process data are acquired in real time, ensuring that the data dimensions meet the requirements of multi-axis collaborative control. A global timestamp clearly marks the acquisition time for each piece of multi-axis raw data, giving the data traceable time attributes and providing a core basis for subsequent time sequence alignment, strengthening the data... Based on the temporal correlation between data points; a real-time upload mechanism ensures the rapid transmission of timestamped data sets to the control cabinet; centralized storage in the data receiving area enables the orderly aggregation of data, avoiding processing delays caused by data transmission delays or scattered storage; time sequence alignment is performed based on global timestamps to accurately eliminate the acquisition time deviation of data on different axes, ensuring that the time-synchronized data sequence maintains consistency in the time dimension and improving the temporal reliability of the data; the multi-axis synchronized raw data stream integrated according to the time axis realizes the unified correlation of data in the time and spatial dimensions, making the spatiotemporal correspondence of data on each axis clear and explicit, providing structured data support for subsequent fusion calculations.

[0020] In a preferred embodiment of the present invention, step 200 involves fusing and calculating the spatial pose data in the multi-axis synchronous raw data stream based on a preset extended kinematic model including the external axis, and performing real-time filtering on the torque and process data to obtain a cooperative state dataset containing the desired state, including: Step 201, receiving the spatiotemporally unified multi-axis synchronous raw data stream, specifically includes: the data receiving module built into the robot's main controller acts as the execution subject, receiving the spatiotemporally unified multi-axis synchronous raw data stream in real time. This data stream is transmitted in the form of a frame sequence through a cooperative motion control network. Each frame of data includes seven parts: frame header, frame sequence number, data length, axis identifier set, timestamp field, data body, and frame tail checksum. The data receiving module first performs frame header recognition on each received frame of data. After confirming that the frame header identifier is consistent with the preset 0xAA55 hexadecimal identifier, it performs CRC-... A 16-cycle redundancy check is performed. If the checksum matches the frame tail field, the frame is considered valid. If the check fails, a retransmission request is sent to the data sender. For valid frames, the data receiving module parses them sequentially according to the frame sequence number, extracts the timestamp field and axis identifier set, and stores the data body according to the axis identifier in the temporary data buffer of the main controller. This buffer adopts a shared memory design, with a read / write rate matching the data stream transmission rate of 100MB / s. At the same time, an independent storage partition is configured for each type of axis data, and the data in the partition is arranged in ascending order of timestamp to ensure the temporal continuity of the data during subsequent processing.

[0021] Step 202: Based on a preset extended kinematics model including external axes, perform forward kinematics fusion calculation on the spatial pose data in the multi-axis synchronous raw data stream to obtain the precise pose of the robot end effector in the global coordinate system. Specifically, this includes: The preset extended kinematics model including external axes needs to complete the construction, training, and solidification process before being pre-placed in the model computation unit of the main controller. Specifically, model construction is based on accurate parameter acquisition, proceeding logically according to the acquisition of basic axis parameters, external axis parameters, and model integration. Firstly, the basic parameters of the robot's six joint axes are acquired. The joint axes are calibrated offline using a laser tracker. Combined with the robot's factory technical documentation, basic DH parameters such as link length, joint offset, and torsion angle are obtained for each joint axis. Simultaneously, the reference values ​​of the joint angles at different rotation angles are recorded through joint idle rotation tests to ensure complete matching between the parameters and the mechanical structure. Secondly, external axis parameters are acquired. For the external track walking axis, the parameters are adjusted on the track installation... In the trial phase, a laser interferometer was used for positioning and calibration to obtain the coordinate offset in the track length direction. Combined with the mechanical design drawings of the walking mechanism and the specifications of the transmission components, linear displacement correlation parameters such as the reduction ratio and transmission efficiency of the walking mechanism were extracted. For the servo positioner axis, the rotation angle limit range of the A-axis and B-axis was recorded through single-axis and dual-axis linkage tests of the positioner. The rotation angle coupling relationship when the two axes move simultaneously was captured by the attitude measurement instrument to determine the linkage compensation coefficient. Thirdly, the model was integrated and constructed. Based on the Denavit-Hartenberg parametric method, a basic kinematic model containing only 6 joint axes was first built with the robot base coordinate system as the origin, and the coordinate transformation relationship between the links of each joint axis was established. Then, the linear displacement parameters of the external track walking axis, the rotation angle parameters of the servo positioner axis, and the linkage compensation coefficient were integrated into the basic model through a homogeneous transformation matrix. The transformation rules between the robot base coordinate system and the global coordinate system were defined to form a complete extended kinematic model including the external axis.

[0022] The core of model training is parameter accuracy verification and iterative correction to ensure that the model's calculation results are consistent with actual working conditions. First, offline simulation training is conducted. The initial model is imported into the robot simulation platform, and 50 sets of typical motion parameters covering joint axes, external track axes, and positioner axes are input. The end-effector pose output under different working postures is simulated, and the deviation between the simulated output pose and the theoretical design pose is compared. If the coordinate deviation exceeds ±0.01mm or the posture angle deviation exceeds ±0.03 degrees, the DH parameters or linkage compensation coefficients of the corresponding axes are adjusted accordingly. Second, physical trial operation training is carried out. The preliminarily corrected model is loaded onto the main controller prototype, controlling the robot to drive the external axes to complete typical tasks such as welding and handling, using a laser tracker. The actual pose data of the robot's end effector is collected in real time and compared point by point with the pose data output by the model. Deviation data at different operation stages are recorded. For example, when the robot travels to the middle position on the external track, the influence of linear displacement parameters on the end effector pose is verified. When the A-axis and B-axis of the positioner are linked, the linkage compensation coefficient of the two axes is corrected. Finally, the parameters are iteratively solidified. The deviation data collected in offline simulation and physical trial operation are statistically analyzed. The least squares method is used to iteratively optimize the model parameters until the coordinate deviation between the model output pose and the actual pose is stable within ±0.01mm and the attitude angle deviation is stable within ±0.01 degrees under 10 consecutive typical working conditions. The model training is completed and the final parameter configuration file is generated.

[0023] The extended kinematics model including the external axis, which has been trained as described above, is fixed in the model computing unit of the main controller in the form of a parameter configuration file. This model is constructed based on the Denavit-Hartenberg parametric method. In addition to the basic DH parameters such as the link length, joint offset, torsion angle, and joint angle of the robot's own six joint axes after verification, it also incorporates the linear displacement parameters of the external track axis, the rotation angle parameters of the servo positioner axis, and the linkage compensation coefficient. The parameters of the external track axis include the coordinate offset in the track length direction and the transmission ratio of the walking mechanism. The parameters of the servo positioner axis include the rotation range of the A-axis and B-axis and the linkage relationship between the two axes; the model calculation unit extracts the spatial pose data of each axis from the temporary data buffer, where the joint axis pose data is the joint rotation angle, the external track walking axis pose data is the linear displacement value, and the servo positioner axis pose data is the real-time rotation angle of the A-axis and B-axis; the global coordinate system is used as the reference, the origin of which is set as the initial position of the external track walking axis, the X-axis is along the length of the track, the Y-axis is perpendicular to the track plane, and the Z-axis is perpendicular to the XY plane. The parameters of this coordinate system have been preset and verified during the model construction stage.

[0024] The model computation unit first calls the DH parameters of the joint axes in the model to perform forward kinematics calculations on each joint axis, obtaining the pose of the robot wrist end effector in the robot's base coordinate system through a homogeneous transformation matrix. Then, it calls the preset base coordinate system to global coordinate system transformation rules in the model, combining the linear displacement data of the external track axes to transform the robot's base coordinate system pose to the global coordinate system. Finally, it calls the linkage logic of the servo positioner axes in the model, incorporating the rotation angle data of the servo positioner axes to correct the influence of workpiece posture changes on the robot's end effector pose through matrix multiplication. The system completes the fusion of multi-axis data. During the calculation process, intermediate results are verified every 0.05ms. The verification method is to compare the current calculation result with the standard deviation range of the same parameters in the model training. If the deviation exceeds the range, the parameter reloading mechanism is triggered to ensure the accuracy of the fusion of parameters of each axis. Finally, the system outputs the precise pose of the robot end effector in the global coordinate system, including the three-dimensional coordinate values ​​of X, Y, and Z and the attitude angles around the X, Y, and Z axes. The accuracy of the coordinate values ​​is controlled within ±0.005mm and the accuracy of the attitude angles is controlled within ±0.001 degrees, which is consistent with the accuracy target of the model training.

[0025] Step 203: Based on the precise pose of the robot end effector in the global coordinate system, a spatial neighborhood with a preset radius is determined centered on it; geometric mean filtering is performed on the torque data in the multi-axis synchronous raw data stream that falls within the spatial neighborhood to filter out high-frequency fluctuations caused by pose micro-changes and noise, and obtain stable load state parameters; synchronously, real-time low-pass filtering is performed on the welding process data in the multi-axis synchronous raw data stream to suppress high-frequency interference and obtain a stable set of process parameters, specifically including: based on the precise pose of the robot end effector, with the spatial point corresponding to the pose as the center of the sphere, the spatial neighborhood with a preset radius is pre-set through the process parameter configuration interface. According to the accuracy requirements of the welding operation, the radius value ranges from 10 to 50 mm, and is set to 20 mm in this embodiment; the model calculation unit calls the spatial retrieval module to traverse all multi-axis torque data in the temporary data buffer area that deviates from the current timestamp of the robot end effector within ±0.1 ms, and determines whether the acquisition position corresponding to each axis torque data falls within the preset spatial neighborhood through spatial coordinate calculation. The acquisition position is obtained by combining the pose data of each axis with the axis installation position parameters; For torque data falling within the spatial neighborhood, outlier removal is first performed. The 3σ criterion is used to determine whether the data is within the normal distribution range, and data exceeding the mean ± 3 times the standard deviation are removed. For the remaining effective torque data, geometric mean filtering is performed. The arithmetic square root of the number of data points after multiplying all effective data is taken to obtain the stable load state parameters within the spatial neighborhood, including the real-time torque, load power, and load change rate of each axis. Simultaneously, for the welding process data in the multi-axis synchronous raw data stream, including welding current, welding voltage, wire feed speed, and shielding gas flow rate, an infinite impulse response low-pass filter is used for real-time filtering. The cutoff frequency of the filter is set to 50 to 200 Hz according to the interference characteristics of the industrial site. In this embodiment, it is set to 100 Hz. The filter coefficients are updated in real time using the recursive least squares method to ensure the filter's adaptive suppression capability against interference of different frequencies. During the filtering process, the slope of the process data change is monitored in real time. When the slope of the change exceeds a preset threshold, the weight of the filter coefficients is automatically increased to improve data stability. Finally, a set of stable process parameters with fluctuation amplitude of less than 5% is output.

[0026] Step 204: Integrate the precise pose, stable load state parameters, and stable process parameter set of the robot end effector in the global coordinate system to obtain a collaborative state dataset containing the desired state. Specifically, this includes: the main controller's dataset integration module calling the data association interface to extract the precise pose, stable load state parameters, and stable process parameter set of the robot end effector in the global coordinate system, using the timestamp as the core association basis to ensure that the three types of data correspond to the multi-axis state at the same operation time; during the integration process, the data types are first standardized, converting the coordinate values ​​of the pose data to floating-point format, the load parameters to integer format, and the process parameters to integer or floating-point format according to their type, and all data are stored in little-endian byte order; subsequently, the data is processed according to the timestamp, precise end effector pose, joint axis load parameters, and external axis load parameters. The data frame structure is constructed sequentially from load parameters and welding process parameters. The end-effector precise pose contains 6 data items, the joint axis load parameters contain 6 data items, the external axis load parameters contain 3 data items, and the welding process parameters contain 4 data items. Each data item is labeled with its unit of measurement and precision level. After integration, the collaborative state dataset undergoes integrity verification to check for missing data or format errors. If missing data is found, it is filled by interpolating similar data from the previous time step. Linear interpolation is used to ensure data continuity. Finally, a collaborative state dataset containing the desired state is generated. The desired state is determined by a baseline model trained using historical best operation data. The dataset is labeled with the deviation range between each parameter and the desired state. This deviation range is pre-configured according to the operation process requirements, providing a clear reference for subsequent control parameter calculations.

[0027] In this embodiment of the invention, frame check and CRC verification ensure the integrity of the received data stream. A dual-port RAM design matches the transmission rate, and data is stored in axis partitions and sorted by timestamp to ensure continuous data sequence and clear attribution. An extended kinematic model incorporates complete external axis parameters, and multi-axis pose fusion is performed based on the global coordinate system. Matrix operations correct for workpiece posture influences, and intermediate result verification improves calculation accuracy, ultimately outputting high-precision end-effector pose data. The filtering range is dynamically determined based on the end-effector pose, and geometric mean filtering removes outliers before calculation, effectively smoothing high-frequency fluctuations in torque data. Low-pass filtering combined with recursive least squares method updates coefficients, adaptively suppressing process data interference and synchronously outputting stable load and process parameters. Timestamps are used as the core to associate precise end-effector pose, stable load parameters, and process parameters, and standardized processing unifies various data formats and storage rules. Integrity checks identify missing data, and interpolation ensures data continuity. The resulting collaborative state dataset has a well-structured and comprehensive information, providing a reliable basis for the accurate calculation of subsequent axis control parameters.

[0028] In a preferred embodiment of the present invention, step 300 above involves synchronously calculating the control parameters of each axis and extracting key point load and pose information based on the cooperative state dataset to obtain a preliminary control parameter set; performing spatial domain load distribution analysis based on the key point load and pose information to obtain load uniformity characteristics; obtaining the load adaptive feedforward gain based on the load uniformity characteristics; and fusing the preliminary control parameter set and the load adaptive feedforward gain to obtain an adaptive multi-axis cooperative motion command set with sequence identifiers, including: Step 301: Receive the cooperative state dataset, calculate the Cartesian space difference between the actual pose and the target pose of each axis, and convert the Cartesian space difference into the trajectory tracking error of each axis using the inverse kinematics model of the extended kinematics model. Specifically, this includes: receiving the cooperative state dataset, which contains real-time collected feedback values ​​for the actual pose data of each axis, and target pose data for each axis derived from a preset task trajectory planning file. This file is pre-stored in the main controller's storage unit and contains the target X, Y, and Z coordinates of each axis in the global coordinate system at each task time, as well as the target attitude angles around the X, Y, and Z axes; the trajectory control module first extracts the actual pose and target pose of each axis corresponding to the same timestamp, and calculates their Cartesian space values ​​respectively. The difference is calculated by subtracting the actual coordinate value from the target coordinate value, and the attitude angle difference is calculated by the minimum angle difference between the target attitude angle and the actual attitude angle, ensuring that the difference range is between -180 degrees and 180 degrees. Then, the extended kinematics inverse model pre-installed in the main controller is called. This model has the same parameters as the extended kinematics model and includes complete kinematic parameters for joint axes, external track walking axes and servo positioner axes. The calculated Cartesian space coordinate difference and attitude angle difference are used as inputs and converted into the corresponding rotation angle error or displacement error of each axis through inverse kinematics operation, that is, the trajectory tracking error of each axis. The joint axis outputs the rotation angle error, the external track walking axis outputs the displacement error, and the servo positioner axis outputs the rotation angle error.

[0029] Step 302: Input the stable load state parameters into the load torque observer to calculate the load dynamic compensation amount. Specifically, the load torque observer is integrated into the torque control module of the main controller and is constructed using an extended Kalman filter structure. Its state equation is pre-set inside the observer, including parameters such as motor rotational inertia, friction coefficient, and transmission efficiency. These parameters are determined through a combination of offline calibration and online correction. During offline calibration, motor characteristic data is collected through no-load test runs, and during online correction, the parameters are dynamically fine-tuned based on real-time operating data. The stable load state parameters obtained in step 203, including the real-time torque, load power, and load change rate of each axis, along with the real-time current and speed data of each axis motor, are input into the load torque observer. The observer first normalizes the input data, converting parameters of different dimensions to the range of 0 to 1. Then, it predicts and estimates the load torque through the state equation, and then corrects the predicted value by combining the motor current feedback data to calculate the load dynamic compensation amount. The compensation amount includes a dynamic friction compensation term that is positively correlated with the motor speed, and an inertial torque compensation term that corresponds to the load change rate. The output frequency of the compensation amount is consistent with the data acquisition period, which is 0.1ms, to ensure real-time response to dynamic load changes.

[0030] Step 303 involves querying the process calibration mapping table based on the stable process parameter set to obtain matching values ​​for the process parameters. Specifically, the process calibration mapping table is pre-stored in the process database of the main controller. This mapping table is constructed using a large amount of orthogonal experimental data. Experimental variables include welding material, plate thickness, weld type, and welding speed. During the experiment, the optimal welding process parameters under different combinations of variables are recorded, including matching values ​​for welding current, voltage, and wire feed speed, as well as corresponding process effect evaluation indicators, such as penetration depth and weld width. The stable process parameter set obtained in step 203 is used as the query keyword and input into the retrieval module of the process database. The retrieval module uses a multi-parameter fuzzy matching algorithm to first match basic process parameters such as welding material and plate thickness, and then fine-tunes the matching results based on the real-time welding speed. When there is a deviation between the welding current, voltage, and other parameters in the stable process parameter set and the standard parameters in the mapping table, the retrieval module calculates the process parameter matching value that best suits the current operating condition through linear interpolation and outputs the matching confidence level. When the confidence level is lower than 90%, a prompt signal is sent to the process monitoring unit.

[0031] Step 304: Finally, the trajectory tracking errors of each axis, the dynamic load compensation, and the process parameter matching values ​​are encapsulated into a unified data structure and integrated to generate a preliminary control parameter set. Specifically, the parameter integration module of the main controller first defines a unified data structure, which includes an axis identifier field, a timestamp field, a trajectory tracking error field, a dynamic load compensation field, a process parameter matching value field, and a data verification field. The trajectory tracking error field contains the specific value and accuracy level of the rotation angle error or displacement error; the dynamic load compensation field contains the sub-item values ​​of the friction compensation and inertia compensation items; and the process parameter matching value field contains the welding current and voltage... The matching values ​​for wire feeding speed are then determined. Subsequently, the tracking errors of each axis obtained in step 301, the dynamic load compensation amount obtained in step 302, and the matching values ​​of process parameters obtained in step 303 are associated and matched according to the axis identifier and timestamp to ensure that the parameters of the same axis at the same time are stored in the same data structure instance. After data filling is completed, each data structure instance is verified by the MD5 message digest algorithm, and a 128-bit check code is generated and stored in the data verification field. Finally, all data structure instances that pass the verification are sorted by axis identifier and integrated to generate a preliminary control parameter set. This parameter set is stored in the form of a linked list for easy subsequent query and modification operations.

[0032] Step 305: Extract the real-time load and pose information of multiple pre-calibrated key position points in the robot's workspace from the collaborative state dataset to obtain key point load-pose data pairs. Specifically, the pre-calibrated key position points are determined by manual teaching with a teach pendant or automatic calibration with a laser tracker. The number is set according to the size of the robot's workspace and the complexity of the operation. In this embodiment, it is set to 20. These key position points are evenly distributed throughout the workspace, especially in areas where the load is prone to sudden changes, such as weld inflection points and workpiece corners. The calibration information of the key position points is stored in the position database of the main controller, including the coordinate value, attitude angle and corresponding calibration number of each key point in the global coordinate system. The key point extraction module of the main controller reads the collaborative state dataset in real time. It compares the real-time pose information in the dataset with the key position point calibration information in the position database using a coordinate matching algorithm. When the deviation between the real-time pose and the calibration coordinate of a certain key position point is less than 5mm and the attitude angle deviation is less than 0.5 degrees, it is determined that the robot end effector is at that key position point. At this time, the real-time load data at that moment is extracted, including the torque and load power of each axis. This data is combined with the corresponding pose information to form key point load-pose data pairs. Each data pair contains a calibration number, timestamp, pose parameters, and load parameters. The extraction frequency is consistent with the update frequency of the collaborative state dataset.

[0033] Step 306: Based on the load data in the keypoint load-pose data pairs, construct a spatial distribution vector of the load values; perform statistical analysis on the spatial distribution vector of the load values ​​to calculate the statistical variance and statistical kurtosis, specifically including: the load analysis module of the main controller first obtains the load data in all keypoint load-pose data pairs, uses the X-axis coordinate of the key position points in the global coordinate system as an index, arranges the load data in index order, and constructs a spatial distribution vector of the load values. The dimension of this vector is consistent with the number of key position points, and each element in the vector corresponds to the load value of a keypoint; then, the spatial distribution vector is analyzed... Statistical analysis is performed by a statistical calculation unit integrated into the load analysis module. When calculating the statistical variance, the arithmetic mean of all load values ​​is first calculated, then the sum of squares of the differences between each load value and the mean is calculated, and finally divided by the number of load values ​​to obtain the statistical variance. This variance reflects the spatial dispersion of the load. When calculating the statistical kurtosis, the arithmetic mean is used as the benchmark, the sum of the fourth power of the differences between each load value and the mean is calculated, and divided by the product of the number of load values ​​and the fourth power of the standard deviation to obtain the statistical kurtosis. This kurtosis reflects the steepness of the load distribution. A kurtosis value greater than 3 indicates the existence of a peak region where the load is concentrated.

[0034] Step 307: Based on the preset uniformity evaluation rules, the statistical variance and statistical kurtosis are used as inputs. A composite index comprehensively characterizing the uniformity of spatial load distribution is obtained by calculating the weighted fusion value of the statistical variance and statistical kurtosis under normalized calibration, serving as the load uniformity feature. Simultaneously, the value of the statistical variance is recorded, and load abrupt change regions are identified based on spatial segments where the statistical kurtosis exceeds a preset kurtosis threshold. Specifically, the preset uniformity evaluation rules are based on the load characteristics of multi-axis collaborative operation of industrial robots, ensuring a high degree of adaptation between the rules and actual working conditions. Firstly, the normalization... The normalized calibration range was determined based on measured load parameter data from the robot's joint axes, external track axes, and positioner axes. Load data was collected covering 10 typical working conditions, including welding, handling, and assembly. The fluctuation range of the load torque was found to be 0 to 85 N·m, corresponding to an actual range of 0 to 92 N·m for the statistical variance. To reserve safety redundancy and cover extreme working conditions, the normalized calibration range of the statistical variance was set to 0 to 100 N·m. Simultaneously, the actual values ​​of the load distribution kurtosis under each working condition ranged from 1.2 to 8.7; therefore, the normalized calibration range of the statistical kurtosis was set to 0 to 10. First, to ensure that the actual data falls completely within the calibration range; second, regarding the allocation of weighting coefficients, an orthogonal experimental design was used to analyze the influence of statistical variance and kurtosis on load uniformity. The experimental results showed that variance contributed 62% to uniformity, and kurtosis contributed 38%. Based on this result, the weight of the normalized statistical variance was set to 0.6, and the weight of the statistical kurtosis was set to 0.4, so that the weighting allocation closely reflects the actual influence; third, to formulate linear transformation rules, in order to make the weighted fusion value have intuitive evaluation significance, a linear transformation formula was designed to map the fusion value to the range of 0 to 10. The transformation coefficients were determined through extreme value calculation. First, the load is set to ensure that the load is absolutely uniform when the fusion value is 0, and extremely uneven when it is 10. Second, the kurtosis threshold is set by collecting kurtosis data under load change conditions and combining it with vibration monitoring results. When the kurtosis exceeds 3.5, the vibration amplitude of the robot end effector will exceed the allowable range of ±0.02mm. Therefore, the preset kurtosis threshold is set to 3.5. Third, the rules are solidified by integrating the above calibration range, weight coefficient, transformation rules and threshold parameters into uniformity evaluation rules. After verification and confirmation by 30 different load distribution conditions, the rules are solidified into the non-volatile storage unit of the main controller evaluation module.

[0035] The aforementioned pre-defined uniformity evaluation rules are pre-installed in the evaluation module of the main controller. These rules include normalized calibration ranges for statistical variance and kurtosis, with the normalized range for statistical variance being 0 to 100 n·m and the normalized range for kurtosis being 0 to 10. First, the normalization logic in the rules is invoked to convert the statistical variance and kurtosis obtained in step 306 to their corresponding normalized ranges. The conversion method is to divide the actual value by the maximum value of the calibration range to obtain dimensionless normalized values. For example, when the actual statistical variance is 50 n·m, the normalized value is 50 ÷ 100 = 0.5; when the actual statistical kurtosis is 5, the normalized value is 5 ÷ 10 = 0.5. Then, according to the pre-defined weight allocation scheme in the rules, a weight of 0.6 is assigned to the normalized statistical variance, and a weight of 0.4 is assigned to the normalized statistical kurtosis. A weighted fusion value is obtained through weighted summation. For example, the fusion value corresponding to the above normalized values ​​is... The sum is 0.5 × 0.6 + 0.5 × 0.4 = 0.5. This fused value is then mapped to the range of 0 to 10 through the preset linear transformation rule in the rules, forming a composite index as a characteristic of load uniformity. The closer the composite index is to 0, the more uniform the load distribution; the closer it is to 10, the more uneven the load distribution. At the same time, the actual value of the statistical variance is recorded according to the rule requirements, which is convenient for subsequent tracing of the original data of the load dispersion. The statistical kurtosis is compared with the preset kurtosis threshold of 3.5 in real time. When the statistical kurtosis of a certain spatial segment exceeds the threshold, the calibration information database of key location points is immediately called. The calibration number of the key point in the spatial segment is matched with the corresponding global coordinate range to clearly identify the specific spatial location of the load mutation area. For example, when the kurtosis of the key point numbered 12 exceeds the standard, the corresponding rectangular area of ​​1500 to 1550 mm on the X-axis and 800 to 850 mm on the Y-axis is identified as the load mutation area.

[0036] Step 308: Based on the load uniformity characteristics, generate an adaptive load feedforward gain according to a preset load-gain mapping rule. Specifically, for non-uniform regions where the statistical variance exceeds a preset variance threshold, generate a trajectory feedforward gain to smooth the velocity curve; for regions with sudden load changes, generate a torque compensation gain to enhance the torque loop. This includes: the preset load-gain mapping rule is based on the load response characteristics of multi-axis collaborative robot operations, combined with control effect verification to determine various parameters and logic; firstly, trigger condition parameter calibration: by collecting load data from 20 typical welding and handling conditions, analyze the correlation between statistical variance and trajectory tracking error. When the statistical variance exceeds 50 N·m, the mean trajectory tracking error increases from ±0.008 mm to ±0.015 mm, therefore the preset variance threshold is set to 50 N·m; simultaneously, through load change simulation experiments, confirm that the amplitude of statistical kurtosis exceeding 3.5 is positively correlated with the vibration amplitude, which is the torque compensation gain. The adjustment provides a basis for the following: First, the matching relationship between gain type and load characteristics is determined. Based on control theory and experimental verification, it is determined that in non-uniform regions (variance exceeding limits), speed control needs to be optimized through trajectory feedforward gain, and in regions with sudden load changes, torque response needs to be enhanced through torque compensation gain, thus forming a corresponding rule between load characteristics and gain type. Second, the gain value range and adjustment rules are formulated. Through orthogonal experiments to test the gain effect under different composite exponents and kurtosis overshoot, the value range of trajectory feedforward gain is determined to be 0.1 to 0.8, with the gain value increasing by 0.08 for every 1 increase in composite exponent; the value range of torque compensation gain is 1.2 to 2.5, with the gain value increasing by 0.3 for every 1 increase in kurtosis overshoot. Third, the rules are verified and solidified. The initial rules are loaded into the controller, and the control effect is tested through 15 sets of different load distribution conditions. After ensuring that the trajectory error and vibration amplitude are within the allowable range, the rules are solidified into the storage unit of the gain configuration module, forming a stable load-gain mapping rule.

[0037] The aforementioned pre-defined load-gain mapping rules are stored in the gain configuration module of the main controller. These rules include the gain type and gain value range corresponding to different load uniformity characteristics. The gain configuration module first calls the data reading logic in the rules to extract the composite exponent, statistical variance value, and load abrupt change region identifier from the load uniformity characteristics, and simultaneously retrieves the preset threshold parameters in the rules. When the statistical variance value exceeds the preset variance threshold of 50 n·m, the current region is determined to be a non-uniform region according to the rules, and the trajectory feedforward gain generation process is triggered. Following the adjustment rule of the composite exponent-gain value in the rules, the larger the composite exponent, the larger the generated trajectory feedforward gain value. This gain is precisely applied to the speed control loop, reducing the impact of load fluctuations on trajectory tracking accuracy by smoothing the acceleration change rate of the speed curve. For example, the gain value generated when the composite exponent is 6 is relatively... When the sum index is 3, the gain is increased by approximately 0.24 to accommodate more pronounced load non-uniformity. When a load abrupt change region is detected, the torque compensation gain generation process is initiated according to rules. By calculating and statistically analyzing the extent to which the kurtosis exceeds the preset threshold of 3.5, the gain value is determined according to the corresponding adjustment ratio in the rules. The greater the excess, the greater the torque compensation gain value. This gain is applied to the torque control loop, which shortens the torque response time and enhances the rapid response capability of the torque output by proportionally increasing the proportional coefficient and integral coefficient of the torque loop. This effectively suppresses the robot end effector vibration caused by load abrupt changes. For example, the gain value when the kurtosis exceeds 2 is approximately 0.3 higher than when the exceedance is 1, in order to cope with more severe load abrupt changes.

[0038] Step 309 involves superimposing and fusing the preliminary control parameter set with the trajectory feedforward gain and torque compensation gain to obtain optimized control commands for each axis. Specifically, this includes: receiving the preliminary control parameter set, trajectory feedforward gain, and torque compensation gain; firstly, verifying the dimensionality of the three types of parameters to ensure that the dimensions of the trajectory feedforward gain match the velocity control parameters in the preliminary control parameter set, and that the dimensions of the torque compensation gain match the torque control parameters in the preliminary control parameter set; after successful verification, using a superposition and fusion algorithm to fuse the parameters, directly superimposing the trajectory feedforward gain onto the corresponding velocity control parameters and the torque compensation gain onto the corresponding torque control parameters. During the superposition process, the fused parameter values ​​are monitored in real time. When a value exceeds a preset safety threshold, it is automatically limited to the safety threshold range to avoid overshoot; after fusion, the optimized control commands for each axis are simulated and verified. By calling the motion simulation model built into the main controller, the axis motion state after command execution is simulated. If the trajectory error and vibration amplitude in the simulation results are within the allowable range, the optimization is deemed effective, and the optimized control commands for each axis are output.

[0039] Step 310 involves attaching a unique sequence identifier, generated based on a global timestamp and the corresponding axis identifier, to each set of optimized axis control commands. This is ultimately integrated into an adaptive multi-axis cooperative motion command set with sequence identifiers. Specifically, this includes: first, extracting the global timestamp and axis identifier corresponding to each optimized axis control command; converting the global timestamp to 32-bit binary code and the axis identifier to 8-bit binary code; combining these two codes bit-by-bit to form a 40-bit basic code; then performing CRC-8 cyclic redundancy check on the basic code to generate an 8-bit checksum; appending the checksum to the end of the basic code to form a 48-bit unique sequence identifier; each sequence identifier corresponds one-to-one with a set of optimized axis control commands; the identifier generation module appends the sequence identifier to the header of the control command to form a complete command frame containing the sequence identifier, axis identifier, control command type, command parameters, and command checksum; finally, arranging all complete command frames for all axes in ascending order of timestamp and integrating them into an adaptive multi-axis cooperative motion command set with sequence identifiers. This command set supports querying, verifying, and retransmitting commands based on sequence identifiers, providing a foundation for subsequent reliable transmission.

[0040] In this embodiment of the invention, the Cartesian space difference calculation fully covers the position and attitude dimensions. An extended kinematic inverse model is used to achieve multi-axis adaptation of the error, ensuring that the converted trajectory tracking error matches the motion characteristics of each axis. The load torque observer optimizes parameters through offline calibration and online correction. First, the input data is normalized to unify the dimensions, then the load dynamic compensation is calculated. The friction and inertia compensation terms are output separately to better reflect the load variation characteristics. The process calibration mapping table is constructed based on a large amount of orthogonal experimental data. Multi-parameter fuzzy matching first matches the basic process parameters and then fine-tunes them. Combined with confidence level prompts, this ensures that the output process parameter matching values ​​are reliable and adaptable to the current working conditions. A unified data structure is used to encapsulate multiple types of parameters, standardizing the data storage format. An exclusive checksum is generated through MD5 verification to ensure integrity. The linked list storage format supports rapid querying and flexible modification of the initial control parameter set. Pre-calibrated key points are densely arranged in load-volatile areas. A coordinate matching algorithm accurately determines whether the robot end effector is in a key position, extracts the corresponding real-time data, and forms a time-series unified key point load-pose data pair. The key position points... As an index, a spatial distribution vector of load values ​​is constructed. Statistical operations are used to obtain the variance reflecting the load dispersion and the kurtosis reflecting the distribution steepness, providing quantitative support for spatial load distribution analysis. Based on preset evaluation rules, the statistical variance and kurtosis are normalized and then weighted and fused to form a composite index that quantifies the uniformity of load distribution. Simultaneously, the variance values ​​are recorded, and spatial segments of load abrupt changes are identified based on preset kurtosis thresholds. Based on load uniformity characteristics and preset mapping rules, an adaptive feedforward gain that precisely matches the load state is generated. The trajectory feedforward gain in non-uniform regions can smooth the velocity curve, and the torque compensation gain in load abrupt change regions can enhance the response capability. The preliminary control parameter set is superimposed and fused with the two types of gains. First, the dimensional consistency of the parameters is checked, and then the fused parameters are limited within a safe threshold to avoid overshoot. Combined with motion simulation to verify the effectiveness of the commands, the reliability of the optimized control commands is improved. A unique sequence identifier is generated based on the global timestamp and axis identifier, attached to each set of optimized commands, and a structurally complete command frame is constructed. These are then sequentially integrated into an adaptive multi-axis cooperative motion command set, facilitating subsequent transmission verification and breakpoint resumption.

[0041] In a preferred embodiment of the present invention, step 400 above, which involves sending the adaptive multi-axis cooperative motion command set to each axis driver via a primary / backup redundant communication link, monitoring the link status and switching the link in case of an anomaly, and executing breakpoint resumption based on the sequence identifier, thereby achieving reliable delivery of the command set, includes: Step 401: Receive the adaptive multi-axis cooperative motion instruction set and initiate the distribution process of the adaptive multi-axis cooperative motion instruction set to each axis driver through the primary redundant communication link, obtaining a distribution task instance carried by the primary link. Specifically, this includes: the instruction distribution module built into the robot main controller acts as the execution body. This module integrates a primary and backup link management unit, pre-configuring the primary redundant communication link as EtherCAT-based and the backup redundant communication link as an optical fiber communication link. The physical interfaces of the two links are independently deployed on different communication boards of the main controller to avoid common-cause failures. After receiving the adaptive multi-axis cooperative motion instruction set with sequence identifiers, the instruction distribution module first performs integrity verification on the instruction set. The sequence identifiers of all instruction packets are traversed to confirm that there are no missing or duplicates. If the verification passes, the main link management unit is activated. The main link management unit reads the unique network address of each axis driver in the instruction set and splits the instruction set into independent single-axis instruction subsets according to the axis identifier. Each subset encapsulates the axis identifier, the total number of instructions, and the transmission priority parameters. The instruction subset corresponding to the welding core axis is set to the highest priority. Then, an instruction issuance start signal is sent to each axis driver through the main redundant communication link. After each axis driver returns a ready response signal, the instruction issuance process is started, and a issuance task instance containing the instruction subset, target axis address, transmission rate, and start timestamp is generated. This instance is stored in the task management buffer of the main controller in real time.

[0042] Step 402: Based on the aforementioned task instance carried by the main link, synchronously monitor the real-time communication quality parameters of the main link. The real-time communication quality parameters include at least the round-trip delay and packet loss rate of data packets. Specifically, the link monitoring module of the main controller is deeply integrated with the physical layer of the main redundant communication link. By embedding the sending timestamp in the frame header of each instruction data packet and extracting the receiving timestamp in the response data packets returned by each axis driver, the time difference between the two is calculated to obtain the round-trip delay of data packets. The delay measurement accuracy is controlled within 1 microsecond. At the same time, the link monitoring module uses 100ms as the statistical period to count the total number of instruction packets sent by the main link and the total number of valid response packets received within this period. The real-time packet loss rate is calculated using the formula (1 - total number of valid response packets / total number of sent packets), where valid response packets must satisfy the condition that the ACK identifier is correct and the checksum is matched. In addition to round-trip delay and packet loss rate, the monitoring module synchronously collects the signal strength and bit error rate parameters of the main link. The bit error rate is calculated by the ratio of the number of CRC check failures in the received data packets to the total number of receptions. All real-time communication quality parameters are stored in the link status database in timestamp order, and the storage period is consistent with the command issuance period, which is 0.1ms, to ensure that the parameters correspond to the timing of the command transmission process.

[0043] Step 403: Based on the real-time communication quality parameters, the link status evaluation function is called to calculate and obtain the link health evaluation value. When the link health evaluation value is lower than the preset health threshold, the main link communication is determined to be abnormal and a link switching instruction is generated. Specifically, the link evaluation unit of the main controller has a pre-built link status evaluation function. The core function of this function is to transform the multi-dimensional real-time communication quality parameters into a single, quantifiable link health evaluation value, providing an intuitive and unified basis for link status determination. The generated evaluation value is mainly used to compare with the preset health threshold to determine whether the main link communication is normal. The function takes the real-time communication quality parameters as input, and first normalizes the round-trip delay, packet loss rate, and bit error rate according to the preset calibration range, converting them into dimensionless values ​​from 0 to 10, eliminating the evaluation bias caused by the difference in dimensions between different parameters. The calibration range of the round-trip delay is 0 to 100ms, the packet loss rate is 0 to 10%, and the bit error rate is 0 to 10. -6 Subsequently, weighting coefficients were assigned to each parameter. Considering the varying sensitivity of command transmission to different communication indicators during the welding process, the round-trip delay weight was set to 0.4, the packet loss rate weight to 0.6, and the bit error rate was used as a correction term. When the bit error rate exceeded 5 × 10⁻⁶, the correction was applied. -7 A penalty of 0.5 is added to the calculation result to highlight the critical impact of packet loss rate on the continuity of instruction transmission.

[0044] The link health evaluation value is calculated by weighted summation, ranging from 0 to 10. A higher value indicates a better link status, and this value directly reflects the link's current ability to carry command transmission. The preset health threshold is set to 6.5 based on the welding process's requirements for the real-time and reliability of command transmission. When the calculated evaluation value is lower than this threshold, it indicates that the link status can no longer meet the transmission requirements of multi-axis collaborative commands. The link evaluation unit immediately determines that the main link communication is abnormal and generates a link switching command containing the abnormality type, current timestamp, and axis identification information to ensure the accuracy and traceability of the abnormality response. At the same time, the abnormal status and evaluation data are recorded in the system log, and the log information is retained for 90 days for subsequent fault analysis and process optimization traceability.

[0045] Step 404: Responding to the link switching command, the command transmission through the main link is stopped, and the communication path is switched to the backup redundant communication link, resulting in a link switching completion status. Specifically, after receiving the link switching command, the link switching unit of the main controller first sends a command transmission stop signal to the main link management unit. This signal contains a forced stop identifier. Upon receiving the signal, the main link management unit immediately stops sending data to the physical layer and marks the uncompleted command packets as pending resumption. Simultaneously, the link switching unit activates the driver module of the backup redundant communication link, sends an initialization command to the backup link, and configures the communication rate and data frame format of the backup link. The parameters, such as verification method, are ensured to be consistent with the main link parameters. After the backup link driver module completes initialization, it sends a link switching notification frame to each axis driver via the fiber optic link. The frame contains the new link identifier and synchronization timestamp. Upon receiving the frame, each axis driver sends back an acknowledgment frame, completing the interactive acknowledgment mechanism for the communication path switching. The entire link switching process is implemented through hardware-level fast switching logic, with the switching time controlled within 5ms to avoid interruption of multi-axis collaborative operation due to link switching. After the switching is completed, the link switching unit generates a link switching completion status, which includes parameters such as switching time and the current status of the backup link, and feeds it back to the command issuing module.

[0046] Step 405: Receive the link switching completion status and, based on the sequence identifier carried in the adaptive multi-axis cooperative motion command set, query the command packet response logs recorded before the main link interruption, fed back by each axis driver, to obtain the identifier of the last successfully responded command packet. Specifically, the main controller's response log management module is responsible for storing the command packet response information of each axis driver. This log adopts a circular buffer structure with a buffer capacity of 1000 records. Each record includes a sequence identifier, axis identifier, response type, response timestamp, and command packet checksum. After receiving the link switching completion status, the response log management module starts the log query process to... The axis identifier in the adaptive multi-axis cooperative motion command set is used as an index to traverse the response logs of the corresponding axis. The query logic is to match the sequence identifier of the command packet with the response type. The successful response type must satisfy that the ACK signal returned by the driver contains a checksum consistent with the command packet, and the response timestamp is before the main link interruption time. By comparing the continuity of the sequence identifiers one by one, the last record marked as a successful response in the same axis driver is located, and the sequence identifier in the record is extracted as the identifier of the last successfully responded command packet. If there is no successful response record for a certain axis, the first command packet identifier of that axis is used as the base identifier to ensure that subsequent transmissions cover the complete command set.

[0047] Step 406: Based on the identifier of the last successfully acknowledged instruction packet, determine its successor instruction packet in the instruction sequence as the start point for resuming transmission. Specifically, this includes: after receiving the identifier of the last successfully acknowledged instruction packet, calling the instruction set index table, which stores the position information of all instruction packets in ascending order using the sequence identifier as the key; querying the position number of the instruction packet corresponding to the identifier in the sequence through the index table, incrementing the position number by 1 to obtain the position number of the successor instruction packet, which is the start point for resuming transmission; and marking the instruction packet corresponding to this start point in the resuming transmission management module. A resume start flag is set in the header of the data frame. The flag uses a 16-bit binary sequence 0xFFFF. The axis identifier, sequence identifier, and data length of the instruction packet are also recorded. If the last successfully acknowledged instruction packet is the last packet of the instruction set for that axis, it is determined that the axis instruction has been completely sent. There is no need to set a resume start point. Only the transmission completion status is recorded. After the resume start points of all axes are determined, the resume management module encapsulates the start point information according to the axis identifier to form a resume task list. The list includes the axis identifier, the start instruction packet identifier, and the number of incomplete instructions.

[0048] Step 407: Based on the start point of the resume transmission, initiate the command packet resume transmission task starting from that point through the backup redundant communication link. Specifically, this includes: sending the resume transmission task list to the backup link management unit; the backup link management unit first performs a communication quality self-check on the backup redundant communication link to confirm that the bit error rate is less than 1×10⁻⁶. -7 Once the transmission delay is less than 20ms, the resume transmission task is initiated. Based on the axis identifier in the resume transmission task list, the corresponding axis's resume transmission start point and all subsequent incomplete instruction packets are reorganized into a resume transmission instruction subset according to the sequence identifier. Each subset header is updated with control information such as the resume transmission identifier, total number of instructions, and start identifier. A polling mechanism based on axis priority is adopted, assigning the highest priority to the resume transmission instruction subset of the welding core axis and initiating its transmission first. The resume transmission instruction subset of the external auxiliary axes is assigned the second highest priority and transmitted sequentially. During the resume transmission process, for each instruction packet transmitted, the backup link management unit records its transmission timestamp and status, while waiting for the driver's response signal to ensure that the instruction packet is traceable.

[0049] Step 408: Real-time monitoring of the status of the resume transmission task, obtaining the count of successfully transmitted command packets. Specifically, this includes: the main controller's resume monitoring module communicating in real time with the backup link management unit, obtaining the real-time status of the resume transmission task by reading the transmission status register of the backup link and the response status register of each axis driver; this status includes the sequence identifier of the currently transmitted command packet, the number of transmitted command packets, the response result of each command packet, etc.; the count of successfully transmitted command packets is obtained by accumulating the number of successfully responded command packets returned by each axis driver, and the count update cycle is consistent with the command transmission cycle, which is 0.1ms; the resume monitoring module compares the count information with the number of incomplete commands in the resume task list in real time, calculates the resume progress of each axis, and feeds back the progress information in real time to the main controller's human-machine interface in the form of a percentage, while storing it in the resume status database; when a command packet is transmitted three times consecutively without receiving a response, the resume monitoring module triggers a retransmission mechanism, with retransmission intervals of 10ms, 20ms, and 50ms respectively. If retransmission fails, fault information is recorded and an alarm is triggered.

[0050] Step 409: When the instruction packet count reaches the total number of packets in the instruction sequence, the instruction set transmission is determined to be complete, and a confirmation status indicating that the instruction set has been completely delivered is generated, thus achieving reliable delivery of the instruction set. Specifically, this includes: the continuation monitoring module continuously comparing the count of successfully delivered instruction packets with the total number of packets in the adaptive multi-axis collaborative motion instruction set. The total number of packets is obtained by traversing the sequence identifiers of the instruction set and summing the number of instruction packets corresponding to different axis identifiers. When the count of successfully delivered instructions for all axes is equal to the total number of instruction packets corresponding to each axis, the instruction set transmission is determined to be complete. At this time, the continuation monitoring module generates a confirmation status indicating that the instruction set has been completely delivered. This status information includes detailed parameters such as the transmission completion time of each axis, the number of successfully delivered instructions, and the number of retransmissions. The confirmation status is first fed back to the core control unit of the main controller. The core control unit associates this status with the real-time operating status of each axis. After confirming that no instructions are missing, it sends an instruction execution start signal to each axis driver. At the same time, the confirmation status is uploaded to the central monitoring system of the production line for production process traceability and equipment operating status statistics, ultimately achieving reliable delivery of the instruction set and meeting the requirements of continuous process for uninterrupted collaboration.

[0051] In this embodiment of the invention, the instruction set is split by axis to improve the targeting of the instruction; the verification before the main link starts ensures the integrity of the instruction; the task instance records key information to provide a basis for subsequent monitoring; core parameters such as round-trip delay and packet loss rate are collected in real time, and the statistical period matches the instruction period to ensure that the parameters correspond to the timing of the transmission process; parameter normalization and weighted calculation ensure that the health assessment is objective, and the preset threshold fits the process requirements to realize the rapid judgment of link anomalies and the generation of instructions; hardware-level switching shortens the time consumption, and the instruction to be continued is marked before switching; the handshake process ensures that each axis switches synchronously and avoids the interruption of collaborative actions; and the log is stored in a structured manner. It facilitates rapid indexing by axis, and a dual matching mechanism ensures accurate identification of successful responses. In the event of no response, the first packet is used as the baseline to ensure complete coverage. The index table quickly locates the command position, with clear starting point markings and categorized status records to avoid duplication or omissions in subsequent transmissions. Backup link self-checks ensure communication quality, and axis-priority delivery improves efficiency. Retransmission command reassembly and status recording facilitate traceability. Real-time reading of the status register ensures accurate counting, with count updates synchronized with the delivery cycle. The retransmission mechanism ensures reliable transmission. Comparison of the count with the total number of packets ensures complete transmission, confirms the status association with each axis information, and uploads it to the monitoring system for easy production process traceability.

[0052] In a preferred embodiment of the present invention, step 500 above involves collecting actual operational feedback data based on a reliably issued instruction set, comparing it with the expected state, calculating the synchronization deviation and response delay, and performing monitoring and alarming. The synchronization deviation and response delay are then used as inputs for the next cycle to drive a new round of collaborative processing. This includes: Step 501, triggering and generating feedback data acquisition instructions based on the reliably issued instruction set, specifically includes: the feedback control module built into the robot main controller as the execution body. This module communicates with the instruction issuing module in real time via an internal bus. When the instruction issuing module generates a confirmation status indicating that the instruction set has been issued, it immediately sends a trigger signal to the feedback control module. The signal contains the total sequence identifier range of the instruction set, the identifiers of each axis, and the instruction execution start timestamp. After receiving the trigger signal, the feedback control module starts the acquisition instruction generation process. First, it reads the feedback acquisition configuration information pre-stored in the process parameter library. This information includes the acquisition type, acquisition cycle, and accuracy requirements of the feedback data for each axis. The acquisition cycle for the joint axis and servo positioner axis is set to 0.1ms, and the acquisition cycle for the external track axis is set to 0.2ms, consistent with the control cycle of each axis. Subsequently, standardized feedback data acquisition commands are generated. The command frame structure includes command identifier, axis identifier, acquisition parameters, synchronization timestamp, and check code. The acquisition parameters clearly indicate the real-time position, speed, and torque data items to be acquired. The synchronization timestamp is consistent with the global timestamp of the command set to ensure the timing correlation between the feedback data and the control commands. After the acquisition command is generated, the feedback control module sends it in parallel to each axis driver through the currently active link in the primary and backup redundant communication links, while recording the sending time and target axis information.

[0053] Step 502: Respond to and execute the feedback data acquisition command, and acquire the real-time position, speed, and torque feedback data of each axis driver after executing the corresponding command to obtain the actual operation feedback dataset. Specifically, this includes: after receiving the acquisition command, the built-in feedback data acquisition unit of each axis driver responds first through a hardware interrupt mechanism and immediately starts the data acquisition operation; position feedback data is acquired through an absolute encoder installed on the motor shaft end, with the resolution of the joint axis encoder set to 17 bits, and the position data of the external track travel axis is acquired through a linear grating ruler; speed feedback data is calculated through the pulse signal frequency of the encoder, using the M / T speed measurement method, counting the number of pulses and the number of high-frequency clocks within a 0.1ms acquisition period, with the calculation accuracy controlled within ±0.1 rpm. Torque feedback data is acquired through a strain gauge torque sensor integrated into the motor output. During the acquisition process, errors caused by sensor temperature drift are filtered out simultaneously. The acquisition unit continuously acquires three types of data according to the acquisition cycle. After each acquisition, the data is encoded into 16-bit binary code. The encoded data, along with the shaft identifier and acquisition timestamp, is encapsulated into a feedback data frame and uploaded in real time to the feedback data receiving buffer of the main controller via an active communication link. The receiving buffer adopts a double buffer design: one buffer is used to receive new data, and the other buffer is used for data reading and processing to avoid data overwriting and reading conflicts. All uploaded feedback data frames must pass CRC-32 verification. If the verification fails, a re-acquisition mechanism is triggered to ensure the integrity and accuracy of the actual operation feedback dataset.

[0054] Step 503: Extract the expected position, velocity, and torque data at the corresponding moment from the cooperative state dataset as the expected state dataset; perform a point-by-point real-time comparison between the actual operation feedback dataset and the expected state dataset to calculate the instantaneous difference in position, velocity, and torque dimensions of each axis, and summarize to generate a state difference dataset. Specifically, this includes: the main controller's data analysis module pre-built a Cartesian space pose deviation geometric quantization algorithm, which can realize the mapping and conversion between joint space and Cartesian space deviation to quantify the geometric deviation between the robot's end-effector's actual pose and the expected pose; the data analysis module first extracts the actual operation feedback dataset from the feedback data receiving buffer, and simultaneously retrieves the corresponding period's cooperative state dataset from the cooperative state data storage unit, using timestamps and axis identifiers. A dual matching mechanism establishes a one-to-one correspondence between two sets of data, ensuring that the comparison is of the operating state and expected state of the same axis at the same time. The expected state dataset extracted from the collaborative state dataset contains the expected position, expected velocity, and expected torque of each axis at the corresponding timestamp. These expected data are jointly determined by the extended kinematic model and the process calibration mapping table, with an accompanying accuracy level label. The data comparison adopts a point-by-point real-time calculation method. The instantaneous difference in the position dimension is obtained by subtracting the actual position from the expected position. Joint axes are represented by angular differences, and external axes are represented by displacement differences. The instantaneous difference in the velocity dimension is calculated by the difference between the expected velocity and the actual velocity, distinguishing the dimensional differences between linear velocity and angular velocity. The instantaneous difference in the torque dimension is obtained by subtracting the actual torque from the expected torque, while also annotating the torque direction information.

[0055] After calculating the instantaneous differences in the joint space and external axis motion space of each axis, the data analysis module initiates the Cartesian space pose deviation geometric quantization algorithm. First, it calls the forward logic of the extended kinematics model to map the actual position difference and the expected position difference of each axis to the Cartesian space of the robot's end effector, obtaining the actual three-dimensional coordinates and the expected three-dimensional coordinates of the end effector. Then, it performs geometric quantization on the position deviation in the Cartesian space, first calculating the single-axis position deviation of the end effector in the X, Y, and Z coordinate axes, and then integrating it into the spatial straight-line distance deviation of the end effector based on spatial geometric relationships, thereby characterizing the overall positional offset of the end effector. Subsequently, it performs geometric quantization on the attitude deviation, converting the angle deviation of each axis into the attitude angle deviation of the end effector around the X, Y, and Z axes, and then obtaining the comprehensive attitude deviation through geometric synthesis, thereby characterizing the overall attitude offset of the end effector. At the same time, for the velocity and torque dimensions, it associates the velocity difference and torque difference of each axis with the motion trend deviation and load deviation in the Cartesian space of the end effector, forming complete deviation data in the Cartesian space.

[0056] After calculating the instantaneous differences and Cartesian space pose geometric quantization deviations for all dimensions, a state difference dataset is generated by summarizing the data according to the axis identifier, timestamp, position difference, velocity difference, torque difference, Cartesian space position deviation, and Cartesian space attitude deviation. Each difference item in the dataset is labeled with its corresponding calculation precision, providing a basis for subsequent deviation analysis.

[0057] Step 504 involves performing time-series analysis on the state difference dataset to calculate the synchronization deviation value and command response delay. Specifically, after receiving the state difference dataset, the timing analysis unit of the main controller first sorts the data in time sequence, arranging all instantaneous differences along the same axis in ascending order of timestamps to form a single-axis difference time-series curve. The synchronization deviation value is calculated using a sliding window method, with a window length of 10 acquisition cycles (1ms). The root mean square of the position and velocity differences within the window is calculated to obtain the average synchronization deviation value within that window. The window slides point by point according to the acquisition cycle to achieve real-time updates of the synchronization deviation. The step deviation value reflects the degree of deviation between the actual trajectory and the expected trajectory during multi-axis motion. The command response delay is calculated by matching the control command issuance timestamp and the feedback data acquisition timestamp. Specifically, the feedback data acquisition timestamp corresponding to the same command is subtracted from the command issuance timestamp to obtain the time interval from the issuance of the command to the generation of the actual motion response. To eliminate random errors, the arithmetic mean of the response delays of three consecutive identical commands is taken as the final command response delay. The timing analysis unit stores the calculated synchronization deviation value and command response delay in association with the axis identifier and timestamp to form a deviation-delay association dataset.

[0058] Step 505: Compare the synchronization deviation value and command response delay with preset thresholds respectively to generate corresponding over-limit status flags, and generate visual alarm signals and levels based on the over-limit status flags. Specifically, this includes: the threshold comparison module of the main controller pre-stores preset thresholds for each axis at different operating stages. These thresholds are determined through a combination of offline debugging and online optimization. The position synchronization deviation threshold is divided according to axis type: ±0.01 degrees for joint axes, ±0.05 mm for external track travel axes, and ±0.02 degrees for servo positioner axes; the speed synchronization deviation threshold is uniformly set to ±1 revolution / minute; the command response delay threshold is set to 20 ms, consistent with the maximum allowable delay required by the process; the threshold comparison module compares each parameter in the deviation-delay correlation dataset with the corresponding preset threshold in real time. When a parameter exceeds the upper limit of the threshold or falls below the lower limit of the threshold, an over-limit status flag corresponding to that parameter is immediately generated. The flag bits use binary encoding, with 1 indicating over-limit and 0 indicating normal. The timestamp of the over-limit occurrence, axis identifier, and over-limit value are recorded simultaneously; an alarm signal is generated based on the over-limit status flag.

[0059] Step 506: The synchronization deviation value and command response delay are used as the feedforward calibration input for the next control cycle to drive a new round of complete collaborative processing, realizing real-time synchronization optimization of multi-axis collaborative data. Specifically, this includes: the feedforward calibration module of the main controller has a pre-built spatial residual geometric correction algorithm. The core function of this algorithm is to quantify the uncompensated deviation residual in Cartesian space through geometric operations, generating a precise correction amount in the spatial dimension, which complements the parameter-level calibration; the algorithm has built-in spatial vector decomposition, rotation matrix adjustment, and residual smoothing processing logic, and its geometric reference is consistent with the global coordinate system to ensure that the correction result is appropriate. The system incorporates multi-axis coordinated spatial motion characteristics. The feedforward calibration module extracts the synchronization deviation value and command response delay from the deviation-delay correlation dataset as the core calibration input parameters for the next control cycle. Simultaneously, it retrieves the Cartesian spatial pose deviation from the state difference dataset generated in step 503 as the initial input data for the spatial residual geometric correction algorithm. First, feature extraction is performed on these parameters to analyze the changing trend of the synchronization deviation and the fluctuation law of the command response delay. If the synchronization deviation exhibits periodic changes, the change period and amplitude are extracted as calibration features. If the command response delay has a fixed offset, the offset is used as the delay compensation feature.

[0060] After feature extraction, the spatial residual geometric correction algorithm is initiated for specific calculations: First, the Cartesian spatial position and attitude deviations are decomposed into linear residuals along the X, Y, and Z axes and rotational residuals around each axis. Vector dot product operations are used to eliminate coupling interference between residuals of different dimensions, ensuring that single-dimensional residual data are independent and reliable. Second, based on the geometric reference of the global coordinate system, spatial distance correction is performed on the linear residuals. Combined with the curvature of the robot's end effector trajectory, the residual values ​​are proportionally adjusted; the greater the curvature, the higher the residual correction weight, adapting to the spatial accuracy requirements of trajectory changes. Third, matrix correction is performed on the rotational residuals. By constructing an attitude correction rotation matrix, the rotational residuals around each axis are converted into joint angle corrections, establishing a direct correlation between spatial attitude deviations and joint parameters. Fourth, the residual correction results for five consecutive control cycles are smoothed using the least squares method to eliminate correction fluctuations caused by instantaneous noise, resulting in stable spatial residual correction features.

[0061] Subsequently, the extracted calibration features and spatial residual correction features are input together into the parameter correction interface of the extended kinematic model. The synchronization deviation value and spatial linear residual correction features are used together to correct DH parameters such as link length and joint offset in the model. The command response delay and spatial rotational residual correction features are used together to adjust the prediction time window of the model, so that the expected state output by the model matches both the parameter characteristics and the spatial geometric motion law. At the same time, the feedforward calibration module sends the synchronization deviation value, command response delay and spatial residual correction features to the load torque observer and the process calibration mapping table. The spatial residual correction features are used to optimize the spatial adaptation coefficient of the load dynamic compensation amount, improve the matching degree between the compensation amount and the end spatial load distribution, and improve the spatial interpolation accuracy of the process parameter matching value, so that the parameter matching is dynamically adjusted with the end spatial position.

[0062] After the calibration parameters are configured, the feedforward calibration module sends a start signal to the main controller core control unit to trigger a new round of collaborative processing, that is, to re-execute the complete logic of steps 100 to 400, realize feedforward self-calibration based on historical operating performance and spatial geometric residuals, so that the synchronization accuracy of multi-axis collaborative data is continuously optimized with the operating cycle, gradually reducing synchronization deviation and command response delay, while reducing the pose residual in Cartesian space, and improving the overall collaborative operation quality and spatial motion accuracy.

[0063] In this embodiment of the invention, feedback acquisition and command execution are linked. Matching acquisition commands with the control cycles of each axis improves targeting, and synchronous timestamps ensure the timing consistency of feedback data and control commands, providing a reliable foundation for subsequent data association. High-precision sensors and professional acquisition methods guarantee data accuracy, temperature drift filtering and verification re-acquisition mechanisms optimize data quality, and a double-buffer design avoids data conflicts, forming a complete and accurate actual operation feedback dataset. A dual-matching mechanism ensures accurate data comparison objects, multi-dimensional difference calculation comprehensively captures operational differences, and a structured summary of state difference datasets provides a clear basis for subsequent analysis. The sliding window method enables real-time dynamic updates of synchronization deviations, averaging multiple sets of data reduces response delay errors, and the associated storage of deviations and delays makes data features more complete. Preset thresholds for each axis and stage align with actual needs, a multi-level alarm mechanism enables abnormal response, and alarm information storage facilitates fault tracing and process optimization. Using deviations and delays as feedforward inputs makes optimization more realistic, correcting model parameters enhances system adaptability, and driving the closed-loop process to continuously improve the multi-axis collaborative data synchronization effect.

[0064] like Figure 2 As shown, embodiments of the present invention also provide a robot control multi-axis collaborative data real-time synchronization optimization system, comprising: The acquisition module is used to establish a cooperative motion control network based on a unified global clock source, synchronously acquire the pose, torque and process data of each axis of the robot and attach a global timestamp to obtain a time-stamped data set; the time-stamped data set is then processed by time alignment to generate a multi-axis synchronous raw data stream; The data processing module is used to perform fusion calculations on the spatial pose data in the multi-axis synchronous raw data stream based on a preset extended kinematic model that includes external axes, and to perform real-time filtering on torque and process data to obtain a cooperative state dataset containing the desired state. The control module is used to synchronously calculate the control parameters of each axis and extract the load and pose information of key points based on the cooperative state dataset to obtain a preliminary control parameter set; combine the load and pose information of key points to perform spatial domain load distribution analysis to obtain load uniformity characteristics; obtain the load adaptive feedforward gain based on the load uniformity characteristics; and fuse the preliminary control parameter set and the load adaptive feedforward gain to obtain an adaptive multi-axis cooperative motion command set with sequence identifiers. The switching module is used to send the adaptive multi-axis cooperative motion command set to each axis driver through the primary and backup redundant communication links, monitor the link status and switch the link when abnormal, and perform breakpoint resume transmission according to the sequence identifier to realize the reliable sending of the command set. The closed-loop verification module is used to collect actual operation feedback data based on the reliably issued instruction set, compare it with the expected state, calculate the synchronization deviation and response delay, and monitor and alarm. The synchronization deviation and response delay are used as the input for the next cycle to drive a new round of collaborative processing.

[0065] It should be noted that this system is a system corresponding to the above method. All implementation methods in the above method embodiments are applicable to this embodiment and can achieve the same technical effect.

[0066] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for real-time synchronization optimization of multi-axis collaborative data in robot control, characterized in that, The method includes: Step 100: Establish a cooperative motion control network based on a unified global clock source, synchronously collect the pose, torque and process data of each axis of the robot and attach a global timestamp to obtain a time-stamped data set; generate a multi-axis synchronous raw data stream by time alignment processing of the time-stamped data set. Step 200: Based on the preset extended kinematic model including the external axis, the spatial pose data in the multi-axis synchronous raw data stream is fused and calculated, and the torque and process data are filtered in real time to obtain a cooperative state dataset containing the desired state. Step 300: Based on the cooperative state dataset, synchronously calculate the control parameters of each axis and extract the load and pose information of key points to obtain a preliminary control parameter set; combine the load and pose information of key points to perform spatial domain load distribution analysis to obtain load uniformity characteristics; obtain the load adaptive feedforward gain based on the load uniformity characteristics; fuse the preliminary control parameter set and the load adaptive feedforward gain to obtain an adaptive multi-axis cooperative motion command set with sequence identifiers. Step 400: The adaptive multi-axis cooperative motion command set is sent to each axis driver through the primary and backup redundant communication links. The link status is monitored and the link is switched when there is an abnormality. The interruption resume is executed according to the sequence identifier to realize the reliable sending of the command set. Step 500: Collect actual operation feedback data based on the reliably issued instruction set, compare it with the expected state, calculate the synchronization deviation and response delay, and monitor and alarm. Use the synchronization deviation and response delay as the input for the next cycle to drive a new round of collaborative processing.

2. The method for real-time synchronization optimization of multi-axis collaborative data in robot control according to claim 1, characterized in that, Step 100 includes: Construct a unified global clock source and establish a cooperative motion control network based on the global clock source. Configure a synchronous clock reference and a unique axis identifier for the sensor data channels corresponding to each axis of the robot in the cooperative motion control network. Through the cooperative motion control network, real-time pose, torque and welding process data of each joint axis of the robot, the external track walking axis and the servo positioner axis are collected synchronously to obtain a multi-axis raw data set; Based on the global clock source, a global timestamp is appended to the multi-axis raw data set to obtain a timestamped data set; The timestamped data set is uploaded in real time to the data receiving area of ​​the robot control cabinet through the cooperative motion control network; The system receives a set of timestamped data from the data receiving area and performs multi-source time-series alignment processing based on the global timestamp carried by each data set to obtain a time-synchronized data sequence. The time-synchronized data sequences are integrated along the time axis to obtain a spatiotemporally unified multi-axis synchronized raw data stream.

3. The method for real-time synchronization optimization of multi-axis collaborative data in robot control according to claim 2, characterized in that, Step 200 includes: Receive the spatiotemporally unified multi-axis synchronous raw data stream; Based on a preset extended kinematics model that includes external axes, the spatial pose data in the multi-axis synchronous raw data stream is fused with forward kinematics to obtain the precise pose of the robot end effector in the global coordinate system. Based on the precise pose of the robot end effector in the global coordinate system, a spatial neighborhood with a preset radius is determined with it as the center; geometric mean filtering is performed on the torque data in the multi-axis synchronous raw data stream falling within the spatial neighborhood to filter out high-frequency fluctuations caused by pose micro-changes and noise, and obtain stable load state parameters; simultaneously, real-time low-pass filtering is performed on the welding process data in the multi-axis synchronous raw data stream to suppress high-frequency interference and obtain a stable set of process parameters. By integrating the precise pose, stable load state parameters, and stable process parameters of the robot end effector in the global coordinate system, a cooperative state dataset containing the desired state is obtained.

4. The method for real-time synchronization optimization of multi-axis collaborative data in robot control according to claim 3, characterized in that, Step 300 includes: Receive the cooperative state dataset, calculate the Cartesian space difference between the actual pose of each axis and the target pose, and convert the Cartesian space difference into the trajectory tracking error of each axis through the inverse kinematics model of the extended kinematics model; Input the stable load state parameters into the load torque observer to calculate the load dynamic compensation amount; Based on the set of stable process parameters, the process calibration mapping table is queried to obtain the matching values ​​of the process parameters; Finally, the trajectory tracking error of each axis, the dynamic compensation amount of the load, and the matching value of the process parameters are encapsulated into a unified data structure and integrated to generate a preliminary control parameter set; From the cooperative state dataset, extract the real-time load and pose information of multiple pre-calibrated key locations within the robot's workspace to obtain key point load-pose data pairs; Based on the load data in the keypoint load-pose data pair, a spatial distribution vector of load values ​​is constructed; statistical analysis is performed on the spatial distribution vector of load values ​​to calculate the statistical variance and statistical kurtosis. Based on the preset uniformity evaluation rules, the statistical variance and statistical kurtosis are used as inputs. By calculating the weighted fusion value of the statistical variance and statistical kurtosis under normalization calibration, a composite index that comprehensively represents the uniformity of spatial load distribution is obtained as a load uniformity feature. At the same time, the value of the statistical variance is recorded, and the load abrupt change region is identified according to the spatial segment where the statistical kurtosis exceeds the preset kurtosis threshold.

5. The method for real-time synchronization optimization of multi-axis collaborative data in robot control according to claim 4, characterized in that, Step 300 also includes: Based on the load uniformity characteristics, a load adaptive feedforward gain is generated according to a preset load-gain mapping rule; wherein, for non-uniform regions where the statistical variance value exceeds a preset variance threshold, a trajectory feedforward gain for smoothing the velocity curve is generated; and for regions with sudden load changes, a torque compensation gain for enhancing the torque loop is generated. The preliminary control parameter set is superimposed and fused with the trajectory feedforward gain and torque compensation gain to obtain the optimized control commands for each axis; Each optimized set of axis control commands is appended with a unique sequence identifier generated based on a global timestamp and the axis identifier corresponding to each axis, and finally integrated into an adaptive multi-axis cooperative motion command set with sequence identifiers.

6. The method for real-time synchronization optimization of multi-axis collaborative data in robot control according to claim 5, characterized in that, Step 400 includes: The adaptive multi-axis cooperative motion instruction set is received, and the process of sending the adaptive multi-axis cooperative motion instruction set to each axis driver is initiated through the main redundant communication link, resulting in a sending task instance carried by the main link. Based on the aforementioned task instance carried by the main link, the real-time communication quality parameters of the main link are monitored synchronously. The real-time communication quality parameters include at least the data packet round-trip delay and packet loss rate. Based on the real-time communication quality parameters, the link status evaluation function is called to calculate and obtain the link health evaluation value; when the link health evaluation value is lower than the preset health threshold, the main link communication is determined to be abnormal and a link switching instruction is generated. In response to the link switching command, the system stops issuing commands through the main link and switches the communication path to the backup redundant communication link, thus obtaining the link switching completion status.

7. The method for real-time synchronization optimization of multi-axis collaborative data in robot control according to claim 6, characterized in that, Step 400 also includes: The system receives the link switching completion status and, based on the sequence identifier carried in the adaptive multi-axis cooperative motion instruction set, queries the instruction packet response log recorded before the main link was interrupted and fed back by each axis driver to obtain the identifier of the last successfully responded instruction packet. Based on the identifier of the last successfully acknowledged instruction packet, the subsequent instruction packet in the instruction sequence is determined as the start point for retransmission. Based on the start point of the resume transmission, the command packet resume transmission and distribution task starting from that point is initiated through the backup redundant communication link; Monitor the status of the resume transmission task in real time and obtain the count of successfully transmitted instruction packets; When the instruction packet count reaches the total number of packets in the instruction sequence, it is determined that the instruction set transmission is complete, and an acknowledgment status indicating that the instruction set has been completely delivered is generated, thereby achieving reliable delivery of the instruction set.

8. The method for real-time synchronization optimization of multi-axis collaborative data in robot control according to claim 7, characterized in that, Step 500 includes: Based on the reliably issued instruction set, a feedback data acquisition instruction is triggered and generated; Responding to and executing the feedback data acquisition command, the real-time position, speed and torque feedback data of each axis driver after executing the corresponding command are collected to obtain the actual operation feedback dataset; Extract the expected position, velocity, and torque data at the corresponding moment from the cooperative state dataset to form the expected state dataset; compare the actual operation feedback dataset with the expected state dataset point by point in real time, calculate the instantaneous difference of each axis in the dimensions of position, velocity, and torque, and summarize to generate a state difference dataset. A time-series analysis was performed on the state difference dataset to calculate the synchronization deviation value and command response delay; The synchronization deviation value and the command response delay are compared with preset thresholds respectively to generate corresponding over-limit status flags, and a visual alarm signal and level are generated based on the over-limit status flags. The synchronization deviation value and command response delay are used as feedforward calibration inputs for the next control cycle to drive a new round of complete collaborative processing, thereby achieving real-time synchronization optimization of multi-axis collaborative data.

9. A robot control multi-axis collaborative data real-time synchronization optimization system, the system implementing the method as described in any one of claims 1 to 8, characterized in that, include: The data acquisition module is used to establish a cooperative motion control network based on a unified global clock source, synchronously acquire the pose, torque and process data of each axis of the robot and attach a global timestamp to obtain a data set with timestamps; The time-stamped data set is time-aligned to generate a multi-axis synchronized raw data stream; The data processing module is used to perform fusion calculations on the spatial pose data in the multi-axis synchronous raw data stream based on a preset extended kinematic model that includes external axes, and to perform real-time filtering on torque and process data to obtain a cooperative state dataset containing the desired state. The control module is used to synchronously calculate the control parameters of each axis and extract the load and pose information of key points based on the cooperative state dataset to obtain a preliminary control parameter set; and to perform spatial domain load distribution analysis by combining the load and pose information of key points to obtain load uniformity characteristics. The load adaptive feedforward gain is obtained based on the load uniformity characteristics; the preliminary control parameter set and the load adaptive feedforward gain are fused to obtain an adaptive multi-axis cooperative motion command set with sequence identifiers. The switching module is used to send the adaptive multi-axis cooperative motion command set to each axis driver through the primary and backup redundant communication links, monitor the link status and switch the link when abnormal, and perform breakpoint resume transmission according to the sequence identifier to realize the reliable sending of the command set. The closed-loop verification module is used to collect actual operation feedback data based on the reliably issued instruction set, compare it with the expected state, calculate the synchronization deviation and response delay, and monitor and alarm. The synchronization deviation and response delay are used as the input for the next cycle to drive a new round of collaborative processing.

Images