Portal type ship deck multi-robot collaborative welding control system

Abstract

The application discloses a gantry type ship deck multi-robot cooperative welding control system and relates to the technical field of automatic welding in ship manufacturing.The gantry type ship deck multi-robot cooperative welding control system comprises a large gantry walking mechanism installed on a factory building track along an X-axis direction, a plurality of small gantry mechanisms, each of which comprises a Y-axis horizontal walking mechanism, a Z-axis lifting mechanism and a theta-axis rotating mechanism and carries a welding robot.The gantry type ship deck multi-robot cooperative welding control system solves the problem that a single point failure in a large welding system leads to the shutdown of the whole line through interlocking hierarchical response mechanism and dynamic decoupling control of a safety domain.When local equipment is abnormal, the system only limits the high-risk actions of the associated area, and the non-associated equipment keeps running at a reduced speed, thereby effectively reducing unnecessary downtime;in combination with cross-domain task migration and long weld intelligent segmentation technology, the welding task can be dynamically redistributed under a fault state, and the production continuity is maximally maintained.

Description

Technical Field

[0001] This invention relates to the field of automated welding technology in shipbuilding, specifically to a gantry-type multi-robot collaborative welding control system for ship decks. Background Technology

[0002] Existing gantry-type multi-robot welding systems suffer from a contradiction between insufficient real-time multi-robot collaboration and system rigidity when welding large 15000×15000mm structural components on ship decks. This manifests as: strong coupling and interlocking limitations: the system requires all robots to simultaneously meet rigid conditions such as origin position, normal limit signals, safe gantry spacing, and fault-free communication before initiating the movement and rotation of the large and small gantry sections; global shutdown due to single-point faults: any robot failing to be in position or experiencing a local malfunction, such as sensor failure or communication delay, will directly freeze the entire production line, rendering the 15m / min high-speed walking mechanism inoperable and severely restricting production efficiency; lack of dynamic task allocation: while the existing CC-LinkIE bus architecture supports multi-PLC linkage, such as communication between the large gantry and four small gantry control cabinets, program scheduling relies on fixed interlocking logic, making it impossible to adjust the robot's working area in real time according to the weld seam distribution, resulting in uneven equipment utilization. Therefore, the urgent problem to be solved is: how to decouple the rigid binding of multi-robot actions while ensuring safety, achieving fault isolation and dynamic collaboration, and avoiding system-wide shutdowns caused by single-point anomalies. Existing technologies struggle to balance the contradiction between highly reliable interlocking and production flexibility, becoming a key obstacle to efficient welding of large ship decks. Summary of the Invention

[0003] To achieve the above objectives, the present invention provides the following technical solution: a gantry-type multi-robot collaborative welding control system for ship decks, comprising:

[0004] The large gantry traveling mechanism is installed on the factory rail along the X-axis.

[0005] Multiple small gantry mechanisms, each including a Y-axis traversing mechanism, a Z-axis lifting mechanism, and a θ-axis rotating mechanism, and equipped with a welding robot;

[0006] The PLC control cabinet group is connected to the large and small gantry control cabinets via CC-LinkIE bus;

[0007] A communication network connects all control units;

[0008] Also includes:

[0009] The safety domain dynamic partitioning module is configured to define an independent safety working domain for each robot using LiDAR and UWB positioning unit.

[0010] The interlocking hierarchical response module is configured to divide interlocking conditions into three levels of response: L1 level triggers local deceleration, L2 level triggers path replanning, and L3 level triggers global shutdown.

[0011] Dynamic task allocation engine, configured to parse weld seam data and allocate robot tasks in real time;

[0012] The digital twin pre-inspection module is configured to perform dual safety verification via a virtual production line before physical actions are performed.

[0013] Preferably, the security domain dynamic partitioning module includes:

[0014] An adjustable radius electronic fence surrounding the robot;

[0015] When a collision risk is detected, the motion restriction unit within the domain freezes only the Z-axis lifting and θ-axis rotation within that safety domain, while the X-axis and Y-axis movements outside the domain continue to run.

[0016] Preferably, the interlocking hierarchical response module executes:

[0017] Level L1 response: When a robot collision warning is issued, the walking mechanism speed is reduced to a preset safe value;

[0018] Level 2 response: When the gantry spacing is less than the dynamic safety threshold, the associated mechanism is suspended and the path replanning algorithm is initiated;

[0019] Level 3 response: Activates the global emergency stop circuit when communication is interrupted or power is abnormal.

[0020] Preferably, the dynamic task allocation engine includes:

[0021] The weld topology parser parses the KCONG data generated by SMARTWELD and divides the dynamic work area;

[0022] The reinforcement learning decision-making unit generates a robot task allocation matrix based on a historical welding efficiency database.

[0023] The conflict resolution unit uses an auction algorithm to allocate weld seam tasks with overlapping work domains.

[0024] Preferably, the dynamic task allocation engine further includes:

[0025] The cross-domain welding protocol unit, when a robot malfunction is detected, migrates its unfinished weld to the nearest reachable robot;

[0026] The long weld seam segmentation unit is used to segment weld seams that exceed the working range of a single machine.

[0027] Preferably, the cross-domain welding protocol unit executes:

[0028] Calculate the Manhattan distance based on spatial coordinates and select the health robot closest to the fault point;

[0029] After verifying the reachability of the target robot's welding torch, the task migration is performed.

[0030] Preferably, the digital twin pre-inspection module includes:

[0031] Kinematic simulation unit to verify joint extreme poses and singularity avoidance paths;

[0032] The physical collision detection unit simulates the interference between the tool's center point and the workpiece;

[0033] The execution authorization unit sends action commands to the physical device only after the dual authentication is successful.

[0034] Preferably, the communication network adopts:

[0035] Time-Sensitive Network Slicing Architecture: Dividing the real-time control channel into a task scheduling channel;

[0036] The real-time control channel transmits robot pose data and limit switch signals.

[0037] The self-organizing network backup channel is automatically activated when the main communication link fails.

[0038] Preferred options also include:

[0039] The dynamic speed reduction controller adjusts the walking motor speed proportionally to the distance when the distance between the two robots approaches a safe threshold.

[0040] An emergency path generator, based on a fast random expanding tree algorithm, plans evacuation paths for faulty robots.

[0041] Preferably, the PLC control cabinet group performs:

[0042] The global binding restriction on X / Y axis motion is removed by using the interlocking hierarchical response module;

[0043] In L1 response state, maintain the large gate movement and small gate movement in the non-associated region.

[0044] This invention provides a gantry-type multi-robot collaborative welding control system for ship decks. It has the following advantages:

[0045] This gantry-type multi-robot collaborative welding control system for ship decks solves the problem of single-point failures causing complete line shutdowns in large welding systems by employing an interlocking hierarchical response mechanism and dynamic decoupling control of the safety domain. When local equipment malfunctions, the system only restricts high-risk actions in the associated area, while non-associated equipment maintains reduced speed, effectively minimizing unnecessary downtime. Combined with cross-domain task migration and intelligent long weld seam segmentation technology, it ensures that welding tasks can be dynamically redistributed under fault conditions, maximizing production continuity.

[0046] This gantry-type multi-robot collaborative welding control system for ship decks, based on a dual pre-inspection mechanism using digital twins and multi-dimensional collaborative control, eliminates equipment collisions and welding quality defects at their source. The cascaded verification of kinematic simulation and physical collision detection precisely intercepts interference risks in path planning; specialized strategies such as dynamic verification of vertical welding height and tolerance control of fillet weld posture effectively ensure the forming quality of critical welds; and a communication redundancy architecture and dynamic deceleration rules further enhance system reliability, achieving synergistic optimization of safety and production efficiency. Attached Figure Description

[0047] Figure 1 This is a schematic diagram of the module interaction of a gantry-type multi-robot collaborative welding control system for ship decks, as described in this invention.

[0048] Figure 2 This is a flowchart illustrating a multi-robot collaborative welding control method for gantry-type ship decks according to the present invention. Detailed Implementation

[0049] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0050] Please see Figure 1 and Figure 2 This invention provides a technical solution: a gantry-type multi-robot collaborative welding control system for ship decks, comprising:

[0051] The large gantry traveling mechanism is installed on the factory rail along the X-axis.

[0052] Multiple small gantry mechanisms, each including a Y-axis traversing mechanism, a Z-axis lifting mechanism, and a θ-axis rotating mechanism, and equipped with a welding robot;

[0053] The PLC control cabinet group is connected to the large and small gantry control cabinets via CC-LinkIE bus;

[0054] A communication network connects all control units;

[0055] Also includes:

[0056] The safety domain dynamic partitioning module is configured to define an independent safety working domain for each robot using LiDAR and UWB positioning unit.

[0057] The interlocking hierarchical response module is configured to divide interlocking conditions into three levels of response: L1 level triggers local deceleration, L2 level triggers path replanning, and L3 level triggers global shutdown.

[0058] Dynamic task allocation engine, configured to parse weld seam data and allocate robot tasks in real time;

[0059] The digital twin pre-inspection module is configured to perform dual safety verification via a virtual production line before physical actions are performed.

[0060] It should be further explained that, during the specific implementation process, the following collaborative control process is executed when the gantry-type ship deck multi-robot collaborative welding control system is running:

[0061] S01. Dynamic division of safety domain: By scanning the working radius of each robot with LiDAR and combining it with UWB positioning data, an independent safety working domain is generated in real time. When an external object is detected to intrude into the safety domain, the Z-axis lifting and θ-axis rotation actions within the domain are immediately frozen, while the X-axis walking of the large gantry and the Y-axis traversing mechanism of the small gantry outside the domain are allowed to continue to operate.

[0062] S02. Interlocking hierarchical response, the response process is as follows:

[0063] Level L1 response: If the robot collision warning system is triggered, the walking speed of the affected robot will be reduced to a preset safe value of 3m / min, while equipment in non-associated areas will continue to operate at its original speed; the robot collision warning system is triggered in cases where the welding torch interferes with the workpiece.

[0064] Level 2 response: When the gantry spacing sensor detects that the spacing is less than the dynamic safety threshold, the action of the associated mechanism is paused and the path replanning algorithm is started. After generating an avoidance path, operation is resumed. The dynamic safety threshold is calculated in real time based on the robot's extended posture.

[0065] Level L3 response: In the event of communication interruption or power failure, the global emergency stop circuit is immediately activated, and all mechanisms cease operation.

[0066] S03. Dynamic Task Allocation: Parse the KCONG weld data generated by SMARTWELD and divide the deck into dynamic work areas according to the welding accessibility model; the reinforcement learning agent outputs the robot-weld optimal matching matrix based on the historical welding efficiency database; when the work areas of multiple robots overlap, the conflict resolution unit uses an auction algorithm to allocate weld tasks, with the highest bidder receiving the task first; among them, the highest bidder is scored based on the remaining workload and distance.

[0067] S04. Fault Tolerance Handling: If a robot reports a fault, such as a servo malfunction, the cross-domain welding protocol unit will migrate its incomplete weld seam to the nearest healthy robot in Manhattan. Before migration, the accessibility of the target robot's welding torch to the target weld seam will be verified. If it is not accessible, the long weld seam segmentation unit will be activated and the weld seam will be split and redistributed according to the preset segmentation rules: each segment shall not exceed 5m.

[0068] S05. Digital Twin Pre-inspection: Before sending physical action instructions, send the action sequence to the virtual production line built in Unity3D: first, perform kinematic simulation to verify whether the joints exceed the limits and whether singularities are avoided; then, use the physics engine to detect potential collisions between the tool center point and the workpiece and equipment; only after both checks pass will the instructions be sent to the physical equipment for execution.

[0069] S06. Communication Assurance: Time-Sensitive Network (TSN) is divided into real-time control slices and task scheduling slices; when the main communication link fails, the Mesh self-organizing network backup channel automatically takes over data transmission.

[0070] S07. Dynamic Cooperative Control: When the distance between two robots approaches the safety threshold, the dynamic deceleration controller adjusts the walking motor speed according to the distance ratio; the faulty robot triggers the emergency path generator, plans a collision-free retreat path based on the RRT algorithm, and prioritizes avoiding the working area of ​​the healthy robot.

[0071] The security domain dynamic partitioning module includes:

[0072] An adjustable radius electronic fence surrounding the robot;

[0073] When a collision risk is detected, the motion restriction unit within the domain freezes only the Z-axis lifting and θ-axis rotation within that safety domain, while the X-axis and Y-axis movements outside the domain continue to run.

[0074] It should be further explained that, during the specific implementation process, the security domain dynamic partitioning module performs the following operations during system operation:

[0075] Dynamic generation of electronic fence: The robot's surrounding environment is scanned by LiDAR, and the three-dimensional coordinates of each robot are obtained in real time by UWB positioning unit. An adjustable columnar safe working area is generated with the robot base as the center, that is, the radius range is 1.5-3m. When the workpiece moves or a temporary obstacle enters the scanning area, the radius of the electronic fence dynamically shrinks or expands according to the preset collision risk model.

[0076] In-domain motion restriction strategy: If the safety domain boundary sensor detects an intrusion risk, such as personnel entering or equipment crossing the boundary, all Z-axis lifting and θ-axis rotation movements within the domain will be immediately frozen, and an L1 level warning signal will be sent to the interlocking graded response module at the same time. The X-axis traveling mechanism of the large gantry and the Y-axis traversing mechanism of the small gantry outside the domain will not be affected and will continue to operate at their original speed. For example, when the safety domain alarm of robot No. 3 is triggered, robots No. 1, 2, and 4 can still continue welding operations.

[0077] Multi-condition handling logic:

[0078] Condition 1: Protrusions on the workpiece surface intrude into the safety zone: When the height of the protrusions identified by the LiDAR is greater than the preset threshold of 50mm, only the robot's Z-axis descent is restricted, while the welding torch posture can still be finely adjusted by rotating the θ-axis.

[0079] Condition 2: Temporary equipment enters the security domain: When the UWB tag triggers an intrusion alarm that lasts for more than 2 seconds, all actions within the domain are frozen until the obstacle is removed and the freeze is automatically lifted.

[0080] Condition 3: Multiple safety zones overlap: When the distance between the safety zones of two robots is less than 1m, the electronic fence automatically shrinks to a minimum radius of 1.5m to avoid accidental freezing.

[0081] The interlocking hierarchical response module executes:

[0082] Level L1 response: When a robot collision warning is issued, the walking mechanism speed is reduced to a preset safe value;

[0083] Level 2 response: When the gantry spacing is less than the dynamic safety threshold, the associated mechanism is suspended and the path replanning algorithm is initiated;

[0084] Level 3 response: Activates the global emergency stop circuit when communication is interrupted or power is abnormal.

[0085] It should be further explained that, in the specific implementation process, the interlocking hierarchical response module executes multi-level responses according to the following logic:

[0086] Level L1 response, i.e., local speed reduction: when the collision warning system detects that the distance between the welding torch and the workpiece is less than the safety margin of 20mm or the safety domain dynamic division module triggers an intrusion alarm, the speed of the associated walking mechanism of the affected robot is immediately reduced to the preset safety value of 3m / min, while the non-associated equipment maintains the original speed.

[0087] Operating condition differentiation: If the alarm source is a sudden change on the workpiece surface, such as a weld protrusion, only the Y / Z axis movement of the current welding robot is restricted; if the alarm source is an external intrusion, such as a tool falling, the movement of all mechanisms within this safety domain is frozen.

[0088] Level 2 response, i.e., path replanning: When the gantry spacing sensor detects that the distance between two gantry points is less than the dynamic safety threshold, the following actions are executed: Pause the associated X-axis movement of the large gantry and the Y-axis lateral movement of the small gantry; activate the potential field method path planning algorithm, using other gantry positions as repulsion sources and the target point as attraction sources, to generate a three-dimensional avoidance path; after path verification, the control mechanism resumes operation along the new path to the target position. The replanning logic prioritizes avoiding stationary obstacles; dynamic obstacles are avoided using predicted trajectories; the dynamic safety threshold is calculated in real-time based on the robot's extension angle: the larger the extension angle, the higher the threshold.

[0089] Level L3 response, i.e., global shutdown: when the communication interruption lasts for more than 500ms or the power supply voltage fluctuates by more than ±15%, the hardware emergency stop circuit is immediately triggered, and all motors are braked; after the emergency stop is released, manual confirmation of reset is required, and the welding task is reinitialized.

[0090] The dynamic task allocation engine includes:

[0091] The weld topology parser parses the KCONG data generated by SMARTWELD and divides the dynamic work area;

[0092] The reinforcement learning decision-making unit generates a robot task allocation matrix based on a historical welding efficiency database.

[0093] The conflict resolution unit uses an auction algorithm to allocate weld seam tasks with overlapping work domains.

[0094] It should be further explained that, in the specific implementation process, the dynamic task allocation engine executes the following collaborative control process:

[0095] The KCONG data generated by SMARTWELD is analyzed to extract the spatial coordinates, length, and welding posture requirements of the weld seam. The working area is dynamically divided according to the robot's welding torch accessibility model: if the weld seam is within 90% of the robot's maximum extension radius, it is included in the core working area; weld seams that exceed this range but meet the segmentation conditions are included in the edge working area and the long weld seam segmentation mark is activated.

[0096] The system calls upon a historical welding efficiency database, which contains matching records of weld leg size, plate thickness, and welding speed, to train a deep Q-network model. When outputting the robot-weld matching matrix in real time, it performs multi-objective optimization, including: Objective 1: Minimize the total robot movement distance; Objective 2: Balance the workload of each robot; Objective 3: Prioritize the allocation of vertical welding tasks to the robot with the largest Z-axis travel margin.

[0097] When the target weld seam work areas of multiple robots overlap, the auction algorithm is activated: each robot submits a bid score based on its remaining workload and weld seam distance; the robot with the highest score wins the weld seam task, where the score = weld seam priority coefficient × (1 - its own workload rate) / distance factor; if the highest score difference is <5%, it is forcibly assigned to a robot with a high historical completion rate.

[0098] The dynamic task allocation engine also includes:

[0099] The cross-domain welding protocol unit, when a robot malfunction is detected, migrates its unfinished weld to the nearest reachable robot;

[0100] The long weld seam segmentation unit is used to segment weld seams that exceed the working range of a single machine.

[0101] It should be further explained that, in the specific implementation process, the cross-domain welding protocol in the dynamic task allocation engine and the long weld seam segmentation unit work together according to the following logic:

[0102] If the robot servo system reports an error for more than 5 seconds, the robot is marked as entering a fault state; its unfinished welds are automatically added to the shared task pool, and the nearest healthy robot is calculated based on Manhattan distance, where the distance calculation only considers the X and Y plane coordinates; the servo system error includes motor overheating or encoder malfunction;

[0103] Check the reachability of the target robot's welding torch posture. If the target weld is located in the vertical welding area of ​​its working envelope, it is necessary to additionally verify whether the remaining Z-axis travel is greater than the weld height plus the safety margin. If the verification fails, long weld segmentation will be triggered.

[0104] For welds that cannot be migrated in one section, segmentation rules are implemented, including: Rule 1: Divide the weld into several equal sections along its length, with each section not exceeding the maximum welding span of a single machine; Rule 2: Avoid weld intersections or corner areas at the segmentation points, and prioritize the midpoint of straight sections; Rule 3: When allocating each segment after segmentation, a mandatory 10% overlap area is added to ensure welding continuity.

[0105] After the task is reassigned, the digital twin pre-inspection module performs dual verification on the migrated weld: kinematic simulation and collision detection. If the pre-inspection fails, it automatically falls back to the shared task pool and initiates a manual intervention notification.

[0106] Cross-domain welding protocol unit execution:

[0107] Calculate the Manhattan distance based on spatial coordinates and select the health robot closest to the fault point;

[0108] After verifying the reachability of the target robot's welding torch, the task migration is performed.

[0109] It should be further explained that, in the specific implementation process, when the cross-domain welding protocol unit performs task migration, the following steps are taken to ensure feasibility:

[0110] Recent robot screening: Based on the spatial coordinates of the faulty robot, i.e., the position in the X and Y planes, calculate the Manhattan distance of all healthy robots: |ΔX|+|ΔY|, and filter the top three candidates with the smallest distance; Key exclusion rule: If the current workload of a candidate robot exceeds a preset threshold, it is automatically removed from the candidate list.

[0111] Welding torch accessibility verification includes: flat welding and vertical welding classification verification; for flat welding seams: check whether the target robot welding torch posture covers the weld seam tilt angle range; for vertical welding seams: additionally verify whether the remaining Z-axis travel is greater than the weld seam height + 200mm safety margin; special handling for corner welding: if the weld seam type is corner welding, it is necessary to simultaneously verify whether the θ-axis rotation angle meets the ±185° range and the anti-collision space.

[0112] After verification, the welding task is bound to the target robot's execution queue; the quality of the first welding arc is monitored in real time, and if an abnormality is detected, the process is immediately paused and a manual review is triggered.

[0113] The digital twin pre-inspection module includes:

[0114] Kinematic simulation unit to verify joint extreme poses and singularity avoidance paths;

[0115] The physical collision detection unit simulates the interference between the tool's center point and the workpiece;

[0116] The execution authorization unit sends action commands to the physical device only after the dual authentication is successful.

[0117] It should be further explained that, in the specific implementation process, the digital twin pre-inspection module achieves dual security verification through the following process:

[0118] Kinematic simulation verification: Receive motion command sequences sent by physical devices and drive the robot digital model in a virtual production line; Check whether the motion path exceeds the limit joint by joint: If any joint angle approaches the hardware limit, automatically trigger the joint space interpolation algorithm to replan the path and avoid singular point areas; Special working condition handling: Add attitude tolerance verification to the fillet welding trajectory to ensure that the welding gun tip always meets the requirement of an angle ≥75° with the normal of the workpiece surface.

[0119] Physical collision detection: Based on the dynamic model of the tool center point TCP, interference checks are performed with high-precision workpiece point cloud data; Static collision detection: Verify whether the poses of the path start point and end point interfere with the workpiece and equipment structure; Dynamic trajectory scanning: Sample along the planned path at 50ms intervals to detect the closest distance between TCP and the environment during the movement.

[0120] Tiered response: When the detection distance is <10mm, it is marked as a high-risk path and replanning is required; when the distance is 10-30mm, the execution speed is reduced to 50% and the path is passed.

[0121] Execution Permission Decision: Only when both kinematic verification and collision detection return a safety flag is an execution command sent to the physical device; if either verification fails, a three-level response is triggered, including the following: First failure: Automatically fine-tune the path and re-perform verification; Second failure: Freeze the task and notify the dynamic task allocation engine to reassign it; Third failure: Escalate to manual intervention, lock the device and push alarm information.

[0122] The communication network adopts:

[0123] Time-Sensitive Network Slicing Architecture: Dividing the real-time control channel into a task scheduling channel;

[0124] The real-time control channel transmits robot pose data and limit switch signals.

[0125] The self-organizing network backup channel is automatically activated when the main communication link fails.

[0126] It should be further explained that, in the specific implementation process, the communication network achieves coordinated control through a Time-Sensitive Networking (TSN) slicing architecture, specifically performing the following operations:

[0127] Dual-channel slicing management: Divide the real-time control channel and set the highest transmission priority; independent task scheduling channel adopts a dynamic bandwidth allocation mechanism: when the real-time channel load exceeds 80%, non-critical task data packets are automatically compressed and log recording is possible; the real-time control channel includes the transmission of robot pose data, limit switch signals and emergency stop commands, while the independent task scheduling channel includes the transmission of weld seam allocation commands and program code.

[0128] The fault switching mechanism includes: the main communication link, i.e., when the fiber optic ring network is interrupted, the Mesh self-organizing network backup channel is automatically activated; the node selection logic is: prioritize connecting to adjacent control cabinets with signal strength greater than the threshold; data transmission continuation guarantee: retransmit the last 3 data packets of the real-time channel at the moment of switching to ensure that pose commands are not lost.

[0129] Multi-condition adaptation strategy: Condition 1, i.e., high real-time demand: welding path adjustment command forcibly preempts the real-time channel, interrupting low-priority data transmission; Condition 2, i.e., batch task issuance: task scheduling channel enables block verification mechanism, and feedback on the receiving status is given every 10% of data packets transmitted. If it fails, the current block is retransmitted; Condition 3, i.e., network congestion: real-time channel triggers traffic shaping, limiting the non-urgent data transmission rate to half.

[0130] Also includes:

[0131] The dynamic speed reduction controller adjusts the walking motor speed proportionally to the distance when the distance between the two robots approaches a safe threshold.

[0132] An emergency path generator, based on a fast random expanding tree algorithm, plans evacuation paths for faulty robots.

[0133] It should be further explained that, in the specific implementation process, the dynamic deceleration controller and the emergency path generator work together to execute the following safety strategies:

[0134] The distance between the two robots is monitored in real time. When the distance approaches the dynamic safety threshold, a proportional deceleration algorithm is activated: for every 10% decrease in the distance from the safety threshold, the walking mechanism speed linearly decreases to the corresponding proportion of the original speed. For example, when the distance is 90% of the threshold, the speed drops to 90%. When the distance reaches the minimum safety value of 2m, the speed is forcibly reduced to the preset safety speed of 1m / min. The dynamic safety threshold increases with the increase of the robot's extension angle. If the associated robot is performing vertical welding during the deceleration process, Z-axis height adjustment is allowed to maintain the welding torch height, while the X / Y axes continue to decelerate.

[0135] Emergency path planning: When the robot triggers a fault freeze, start the retreat path planning based on the RRT algorithm: generate an initial path tree with the fault point as the starting point and the nearest safe docking area as the ending point; force path nodes to avoid the working domain of healthy robots; optimize the straightening of path segments to ensure the shortest total length and turning angle ≤45°; where fault freeze includes servo overheating.

[0136] Multi-obstacle response: Static obstacles use a fixed avoidance radius; dynamic obstacles predict their trajectory and allow a 3-second safety time difference.

[0137] Path execution guarantee: The planned path is sent to the digital twin pre-inspection module for collision verification; after the verification is successful, the associated area is locked, and the faulty robot is controlled to retreat along the path to the docking area, during which an audible and visual warning is triggered.

[0138] PLC control cabinet group execution:

[0139] The global binding restriction on X / Y axis motion is removed by using the interlocking hierarchical response module;

[0140] In L1 response state, maintain the large gate movement and small gate movement in the non-associated region.

[0141] It should be further explained that, in the specific implementation process, the PLC control cabinet group removes the global binding restriction of multiple mechanism actions through the interlocking hierarchical response module, and specifically executes the following coordinated control:

[0142] Interlocking signal classification and processing: Receive alarm signals from the safety domain dynamic partitioning module, welding robot, and sensors, and classify them into L1, L2, and L3 level responses according to preset rules; among them, L1 level signals only mark the coordinates of the fault source; L2 level signals are accompanied by a list of associated device IDs; L3 level signals directly trigger the global emergency stop circuit; L1 level signals include single robot collision warnings, L2 level signals include insufficient gantry spacing, and L3 level signals include bus communication interruptions.

[0143] Tiered response execution: L1 level response status: Freeze the Z-axis lifting and θ-axis rotation movements within the safety domain corresponding to the alarm coordinates; maintain the original speed operation of the X-axis travel mechanism of the large gantry and the Y-axis traverse mechanism of the small gantry in the non-associated area; limit the speed of the travel mechanism associated with the fault source to the preset safety value of 3m / min. L2 level response status: Suspend the gantry mechanism associated with the alarm signal; non-associated gantry continue to operate, but the upper limit of travel speed is reduced to 10m / min; automatically release the suspension of associated mechanisms after path replanning is completed.

[0144] Security Mechanism: During L1 / L2 response, the operating status of non-associated devices is monitored in real time. If a new alarm is triggered in a non-associated area, the response level is immediately upgraded to L2 or L3. If the fault source is restored, the restriction is lifted after a 5-second confirmation delay. All response operation logs are synchronized to the digital twin pre-inspection module in real time for post-event analysis.

[0145] Further explanation is needed regarding the overall system operation process during implementation: After the gantry-type ship deck multi-robot collaborative welding control system is started, it first scans the working area of ​​each robot using LiDAR and generates an independent safe working domain by combining ultra-wideband positioning data. The safe working domain is a cylindrical space surrounding the robot base, and its radius can be configured according to the workshop environment. When an external object is detected intruding into the boundary of the safe domain, the movement of the lifting and rotating mechanisms within the domain is immediately frozen, while the walking and traversing mechanisms outside the domain maintain their original speed. The interlocking hierarchical response module receives various alarm signals in real time: if a collision warning or safe domain intrusion is triggered, it is marked as a Level 1 response, which only restricts the movement speed of the mechanism associated with the alarm source; if the gantry spacing sensor detects that the spacing is lower than the dynamic safety threshold, it is marked as a Level 2 response, which suspends the associated gantry and starts the path replanning algorithm; if communication is interrupted or the power supply is abnormal, a Level 3 response global shutdown is immediately triggered.

[0146] The dynamic task allocation engine analyzes the weld data generated by the welding software and dynamically divides the work area based on the robot's joint limit angles and lifting stroke. Welds in the core work area are directly assigned to the optimal robot by a reinforcement learning agent, with simultaneous optimization of movement distance and workload balancing during allocation. When multiple robot work areas overlap, the conflict resolution unit uses an auction algorithm to allocate tasks, with bidding and scoring comprehensively considering weld priority, robot remaining load, and spatial distance. If a robot malfunctions, the cross-domain welding protocol automatically migrates its unfinished welds to the nearest healthy robot in Manhattan. Before migration, the accessibility of the target robot's welding torch to the weld must be verified. For vertical welds, it is necessary to additionally confirm that the lifting stroke margin is greater than the weld height plus a safety margin; for fillet welds, the rotation angle range and lateral collision avoidance space must be verified. For long, inaccessible welds, the system automatically divides them by the midpoint of the straight segment and adds an overlapping area.

[0147] All motion commands must undergo dual verification by the digital twin pre-inspection module before execution. The kinematic simulation unit drives the virtual robot model, checking if joint angles are approaching hardware limits and automatically avoiding singularity trajectories. The physical collision detection unit scans for interference risks throughout the path based on the welding torch's 3D model and workpiece point cloud data. Commands are only issued to the physical device after both verifications pass. The communication network employs a dual-channel architecture: the real-time control channel transmits pose and emergency stop signals with the highest priority; the task scheduling channel dynamically allocates bandwidth to transmit the welding program. In the event of a primary communication failure, the self-organizing network backup channel automatically activates, prioritizing node switching based on signal strength and retransmitting critical data packets.

[0148] Interlocking hierarchical response implementation: In Level 1 response, the system only freezes lifting and rotation movements within the safety domain corresponding to the alarm coordinates. Walking and traversing mechanisms in non-associated areas maintain their original speeds, while the speed of walking mechanisms associated with the fault source is limited to a preset safety value. If the alarm source is a sudden change on the workpiece surface, only the robot's descent movement is restricted, allowing for minor attitude adjustments; if it is an external intrusion, all movements within the domain are frozen. When a Level 2 response is triggered, the movements of associated gantry mechanisms are suspended, and non-associated gantry mechanisms decelerate. The path replanning algorithm prioritizes bypassing static obstacles, predicts trajectories for dynamic obstacles, and calculates safe intersection points. After fault recovery, confirmation is delayed before lifting restrictions to avoid momentary false alarms.

[0149] Cross-domain welding protocol implementation: After a faulty robot is marked, the system filters for three healthy robot candidates closest to the Manhattan distance. If a candidate's current load exceeds a specific multiple of its historical average, it is automatically removed from the candidate list. During task migration, classification verification is performed: for flat welds, the welding torch tilt angle coverage is checked; for vertical welds, the lifting stroke margin is confirmed to be greater than the weld height plus a fixed safety margin; for fillet welds, the rotation angle limits and lateral clearance are additionally checked. If verification fails, long weld segmentation is triggered: the weld length is divided equally, with segmentation points avoiding intersections and corner areas. After segmentation, each segment is forcibly fitted with an overlapping area to ensure welding continuity. Migration tasks must pass digital twin pre-inspection before execution; if the initial welding arc quality is abnormal, the process is immediately paused and transferred to manual review.

[0150] Digital twin pre-inspection implementation: When verifying joint motion paths through kinematic simulation, automatic interpolation is used to generate avoidance paths for joints approaching hardware limits. The fillet weld trajectory must additionally meet the minimum tolerance of the angle between the welding torch tip and the workpiece surface normal. Physical collision detection is performed in two levels: static detection verifies the pose interference of the start and end points, while dynamic scanning samples path points at fixed time intervals. When the detection distance is less than a critical value, high-risk paths are deemed high-risk and require replanning; medium-risk paths are allowed to proceed at reduced speed. Verification failure triggers a three-level process: the first failure automatically fine-tunes the path offset and retryes; the second failure notifies the task engine for reassignment; the third failure locks the device and sends an alarm.

[0151] Dynamic collaborative control implementation: When the distance between two robots approaches the dynamic safety threshold, the walking mechanism speed decreases linearly according to the distance reduction ratio. At the minimum safe distance, it is forcibly reduced to a preset safe speed. If the associated robot is performing vertical welding, the lifting mechanism is allowed to fine-tune to maintain the welding torch height. Faulty robot triggers retreat path planning: Starting from the fault point and ending at the safe stopping area, path nodes are forced to avoid the healthy robot's working area and the avoidance radius is expanded. Path straightening ensures that turning angles do not exceed a fixed angle. Static obstacles are avoided with a fixed radius, and dynamic obstacles have a safety time difference. The planned path must be verified by digital twin collision; after verification, the associated area is locked and retreat is executed.

[0152] Dynamic adjustment of safety domains: The radius of the electronic fence is dynamically adjusted according to the type of obstacle; protrusions on the workpiece surface trigger partial action restrictions based on height thresholds; temporary equipment carrying positioning tags freezes all actions when it continuously intrudes; when multiple safety domains overlap, they automatically shrink to the minimum radius to reduce false alarms.

[0153] Communication fault handling: Real-time control automatically compresses non-critical data packets when the channel load is too high. Batch task distribution employs a block verification mechanism, confirming reception status after transmitting a fixed proportion of data packets. Real-time channel traffic shaping is triggered during network congestion to limit the transmission rate of non-urgent data.

[0154] System recovery mechanism: Manual reset and initialization are required after the global emergency stop is lifted. Task interruption points are automatically saved, and the task resumes from the most recent verification point upon recovery. Interlock response logs are synchronized to the digital twin system, supporting backtracking analysis of fault scenarios.

[0155] It should be further explained that, in the specific implementation process, a multi-robot collaborative welding control method for gantry-type ship decks includes the following steps:

[0156] Step S1: Scan the robot's working environment with LiDAR and generate an independent columnar safe working area in real time by combining ultra-wideband positioning data; when an external object is detected to intrude into the boundary of the safe working area, immediately freeze the lifting and rotating mechanisms within the area, while the walking and traversing mechanisms outside the area maintain their original speed.

[0157] Step S2: Receive and classify sensor alarm signals: Collision warning or security domain intrusion triggers Level 1 response, which only limits the speed of the alarm source associated mechanism; Insufficient gantry spacing triggers Level 2 response, which suspends the associated gantry mechanism and initiates path replanning; Communication interruption or power failure triggers Level 3 response, which shuts down the entire system.

[0158] Step S3: Analyze welding data and divide the work area according to robot accessibility; use reinforcement learning agents to optimize the allocation of core area welds; use auction algorithm to allocate welds in case of conflict; when the robot fails, select the nearest healthy robot and verify the accessibility of the welding torch: for vertical welding, it is necessary to confirm that the lifting stroke margin is greater than the weld height plus the fixed safety margin; for fillet welding, check the rotation angle and lateral gap; unreachable welds are automatically segmented and overlapping areas are added.

[0159] Step S4: Kinematic simulation verifies joint limits and singularity avoidance; fillet welds additionally meet the normal angle requirements; physical collision detection scans the entire path to detect interference risks: static detection of start and end points, dynamic sampling at fixed intervals; instructions are issued only after dual verification is passed.

[0160] Step S5: Communication redundancy guarantee: The real-time control channel prioritizes the transmission of pose and emergency stop commands; the task scheduling channel dynamically allocates bandwidth; in the event of a main communication failure, the self-organizing network backup channel selects nodes based on signal strength and retransmits critical data packets.

[0161] Step S6: When the robot spacing approaches the dynamic safety threshold, the walking mechanism speed decreases linearly according to the spacing reduction ratio; the faulty robot triggers retreat path planning: avoids the working area of ​​the healthy robot and expands the avoidance radius, avoids static obstacles with a fixed radius, and reserves a safety time difference for dynamic obstacles; the planned path is executed after being verified by digital twin.

[0162] Step S7: After the Level 1 response fault is cleared, confirm the delay before resuming the action; after the Level 2 response path replanning is completed, the pause is automatically lifted; all response logs are synchronized to the digital twin system for analysis.

[0163] By employing an interlocking hierarchical response mechanism and dynamic decoupling control of the safety domain, the system solves the problem of single-point failures causing complete line shutdowns in large welding systems. When local equipment malfunctions, the system only restricts high-risk actions in the associated areas, while non-associated equipment maintains reduced speed operation, effectively minimizing unnecessary downtime. Combined with cross-domain task migration and intelligent long weld seam segmentation technology, the system ensures that welding tasks can be dynamically redistributed under fault conditions, maximizing production continuity.

[0164] A dual pre-inspection mechanism based on digital twins and multi-dimensional collaborative control eliminates equipment collisions and welding quality defects at their source. The cascaded verification of kinematic simulation and physical collision detection accurately intercepts interference risks in path planning; specialized strategies such as dynamic verification of vertical welding height and tolerance control of fillet weld posture effectively ensure the forming quality of critical welds; and a communication redundancy architecture and dynamic deceleration rules further enhance system reliability, achieving synergistic optimization of safety and production efficiency.

[0165] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0166] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A gantry-type multi-robot collaborative welding control system for ship decks, comprising: The large gantry traveling mechanism is installed on the factory rail along the X-axis. Multiple small gantry mechanisms, each including a Y-axis traversing mechanism, a Z-axis lifting mechanism, and a θ-axis rotating mechanism, and equipped with a welding robot; The PLC control cabinet group is connected to the large and small gantry control cabinets via CC-LinkIE bus; A communication network connects all control units; Its characteristic is that it further includes: The safety domain dynamic partitioning module is configured to define an independent safety working domain for each robot using LiDAR and UWB positioning unit. The interlocking hierarchical response module is configured to classify interlocking conditions into three levels of response: L1 level response triggers local deceleration, L2 level response triggers path replanning, and L3 level response triggers global shutdown; the interlocking hierarchical response module executes: Level L1 response: When a robot collision warning is issued, the walking mechanism speed is reduced to a preset safe value; Level 2 response: When the gantry spacing is less than the dynamic safety threshold, the associated mechanism is suspended and the path replanning algorithm is initiated; Level L3 response: Activates the global emergency stop circuit when communication is interrupted or power is abnormal; A dynamic task allocation engine is configured to parse weld seam data and allocate robot tasks in real time; the dynamic task allocation engine includes: The weld topology parser parses the KCONG data generated by SMARTWELD and divides the dynamic work area; The reinforcement learning decision-making unit generates a robot task allocation matrix based on a historical welding efficiency database. The conflict resolution unit uses an auction algorithm to allocate weld seam tasks with overlapping work domains; The dynamic task allocation engine dynamically divides the work area based on the robot's joint limit angles and lifting stroke. If the weld is within 90% of the robot's maximum extension radius, it is included in the core work area; welds that exceed this range but meet the segmentation conditions are included in the edge work area and the long weld segmentation mark is activated; welds in the core work area are directly allocated to the optimal robot by the reinforcement learning agent, and the movement distance and workload balance are optimized simultaneously during allocation. The digital twin pre-inspection module is configured to perform dual safety verification via a virtual production line before physical actions are performed.

2. The gantry-type multi-robot collaborative welding control system for ship decks according to claim 1, characterized in that: The security domain dynamic partitioning module includes: An adjustable radius electronic fence surrounding the robot; When a collision risk is detected, the motion restriction unit within the domain freezes only the Z-axis lifting and θ-axis rotation within that safety domain, while the X-axis and Y-axis movements outside the domain continue to run.

3. A gantry-type multi-robot collaborative welding control system for ship decks according to claim 2, characterized in that: The dynamic task allocation engine also includes: The cross-domain welding protocol unit, when a robot malfunction is detected, migrates its unfinished weld to the nearest reachable robot; The long weld seam segmentation unit is used to segment weld seams that exceed the working range of a single machine.

4. A gantry-type multi-robot collaborative welding control system for ship decks according to claim 3, characterized in that: The cross-domain welding protocol unit executes: Calculate the Manhattan distance based on spatial coordinates and select the health robot closest to the fault point; After verifying the reachability of the target robot's welding torch, the task migration is performed.

5. A gantry-type multi-robot collaborative welding control system for ship decks according to claim 1, characterized in that: The digital twin pre-inspection module includes: Kinematic simulation unit to verify joint extreme poses and singularity avoidance paths; The physical collision detection unit simulates the interference between the tool's center point and the workpiece; The execution authorization unit sends action commands to the physical device only after the dual authentication is successful.

6. A gantry-type multi-robot collaborative welding control system for ship decks according to claim 1, characterized in that: The communication network adopts: Time-Sensitive Network Slicing Architecture: Dividing the real-time control channel into a task scheduling channel; The real-time control channel transmits robot pose data and limit switch signals. The self-organizing network backup channel is automatically activated when the main communication link fails.

7. A gantry-type multi-robot collaborative welding control system for ship decks according to claim 1, characterized in that: Also includes: The dynamic speed reduction controller adjusts the walking motor speed proportionally to the distance when the distance between the two robots approaches a safe threshold. An emergency path generator, based on a fast random expanding tree algorithm, plans evacuation paths for faulty robots.

8. A gantry-type multi-robot collaborative welding control system for ship decks according to claim 1, characterized in that: The PLC control cabinet group performs: The global binding restriction on X / Y axis motion is removed by using the interlocking hierarchical response module; In L1 response state, maintain the large gate movement and small gate movement in the non-associated region.

Images