Three-gun welding equipment cooperative real-time process control system

Abstract

This invention belongs to the field of welding automation control, specifically relating to a collaborative real-time process control system for a three-gun welding equipment. The system includes: a physical sensing layer for real-time monitoring of the current vector, spatial pose, magnetic induction intensity distribution, and temperature gradient of the three welding torches; a spatial intelligent decoupling layer, deploying an active dynamic magnetic field neutralization module to generate an anti-phase electromagnetic field, aiming to counteract the mutual inductance magnetic field interference between adjacent welding torches; a collaborative decision-making layer, utilizing a biomimetic cluster intelligent controller, based on thermal flow field topology analysis and optimization strategies, to calculate optimal electrical parameters, path compensation, and magnetic field neutralization parameters in real time; and an execution layer for driving the three welding torches to perform synchronous welding and motion status feedback. Through the above technical solutions, this invention can effectively eliminate electromagnetic interference during multi-gun operation, suppress thermal stress deformation, and significantly improve production efficiency and intelligence level while ensuring weld quality and consistency.

Description

Technical Field

[0001] This invention belongs to the field of welding automation control, specifically relating to a collaborative real-time process control system based on three-gun welding equipment. Background Technology

[0002] With the evolution of intelligent manufacturing technology, multi-station and multi-robot collaborative welding systems are playing an increasingly important role in the processing of large and complex structural components. The dual pursuit of production efficiency and welding quality in industrial manufacturing has driven the transformation of welding equipment from single-gun operation to multi-gun synchronous operation, aiming to shorten processing cycles and reduce energy consumption. In the precision manufacturing of automobile chassis and heavy bridge components, the data processing accuracy and real-time process control capabilities of multi-gun collaborative systems determine the mechanical properties and structural stability of the final product.

[0003] The coordinated process control of a three-gun welding system is the core element for achieving high-efficiency production. This technology achieves multi-point synchronous molten pool construction on the same workpiece surface through the coordinated scheduling of three welding trajectories, current parameters, and spatial posture. This operating mode requires the controller to have extremely high logic analysis capabilities to coordinate the synchronous relationship between multiple heat inputs and complex mechanical movements. Its basic goal is to maximize the deposition efficiency per unit time while ensuring the stability of the arc of each welding gun.

[0004] Existing technologies generally employ strategies such as time-slice polling or serial trajectory scheduling, attempting to avoid magnetic blow interference between multiple arcs by artificially increasing welding gaps or staggering operations. This passive avoidance control mode results in a significant amount of non-productive waiting time during production, limiting the improvement of overall operational efficiency. Simultaneously, existing thermal field control models lack pre-compensation mechanisms for the dynamic coupling of multi-source heat fluxes, making it difficult to effectively suppress component deformation caused by thermal stress concentration under synchronous welding conditions. Because the system lacks proactive feedback adjustment and spatial magnetic field neutralization mechanisms for the real-time changing electromagnetic field environment, the weld droplet transfer process is susceptible to random interference from current fluctuations in adjacent welding torches, resulting in welding quality that fails to meet high-precision process requirements. Summary of the Invention

[0005] The purpose of this invention is to provide a collaborative real-time process control system based on a three-gun welding equipment, which can solve the problems of mutual interference of multiple electric arc magnetic fields, heat accumulation deformation, and difficulty in balancing welding efficiency and quality in the above-mentioned background technology.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows: The collaborative real-time process control system based on three-gun welding equipment includes a physical sensing layer, a spatial intelligent decoupling layer, a collaborative decision-making layer, and an execution layer, wherein: The physical sensing layer includes an electromagnetic field and thermal field fusion sensor array, which is used to monitor in real time the dynamic welding current vector, spatial pose coordinates, local magnetic induction intensity distribution of the welding area and the real-time three-dimensional temperature gradient of the workpiece surface generated by the three welding torches, and to convert the collected physical quantity information into standardized digital signals and transmit them to the collaborative decision layer in real time. The spatial intelligent decoupling layer includes an active dynamic magnetic field neutralization module, which is distributed in the outer periphery of each welding torch. It generates an anti-phase electromagnetic field according to the control command output by the collaborative decision layer to cancel the mutual inductance magnetic field interference between adjacent welding torches, thereby achieving independent purification of the welding arc environment in the spatial dimension. The collaborative decision-making layer includes a biomimetic cluster intelligent controller, which receives multi-dimensional sensor data fed back by the physical perception layer. Based on the thermal flow field topology analysis algorithm and the biomimetic cluster optimization strategy, it calculates in real time the optimal electrical parameter combination of each welding torch, the spatial path compensation amount, and the magnetic field neutralization parameter of the spatial intelligent decoupling layer, thereby realizing nonlinear closed-loop control of the entire welding process. The execution layer includes three independent servo drivers for welding torches and corresponding welding execution mechanisms. The three independent servo drivers are used to drive the three welding torches to perform synchronous welding operations according to the instructions issued by the collaborative decision-making layer, and to provide real-time feedback on the motion status of each drive axis to the collaborative decision-making layer.

[0007] Preferably, the electromagnetic field and thermal field fusion sensor array includes multiple high-frequency Hall sensors, a multispectral thermal imaging device, and a high-precision displacement monitoring unit; the high-frequency Hall sensors are installed in a preset array distribution around the weld pool to capture transient magnetic field changes caused by welding current fluctuations in real time; the multispectral thermal imaging device is used to obtain the thermal radiation intensity distribution on the workpiece surface and construct a dynamic three-dimensional thermal field model.

[0008] Furthermore, the active dynamic magnetic field neutralization module includes a miniaturized Helmholtz coil array and a matching high-speed, high-current power drive unit. The Helmholtz coil array surrounds the nozzle of the welding torch. The high-speed, high-current power drive unit injects a compensation current that is opposite in phase and matches the amplitude of the interfering magnetic field into the Helmholtz coil array according to the pulse signal from the collaborative decision layer, thereby generating a locally controlled canceling magnetic field in the welding torch working area, so that the arc shape of each of the three welding torches remains stable when they are working simultaneously.

[0009] Furthermore, the collaborative decision-making layer incorporates pre-compensation logic based on thermal flow field topology analysis. This pre-compensation logic analyzes the real-time temperature distribution data uploaded by the physical sensing layer to identify the heat accumulation center and thermal flow field evolution trend on the workpiece surface, and dynamically adjusts the instantaneous energy input of each welding torch accordingly. When the predicted thermal stress in a specific area exceeds a preset threshold, the collaborative decision-making layer reduces the welding current of the corresponding welding torch in that area and simultaneously increases the travel speed of the other two welding torches to ensure the dynamic balance of the overall thermal field.

[0010] Furthermore, the biomimetic cluster intelligent controller employs an improved particle swarm optimization algorithm for real-time parameter decision-making. It maps the current, voltage, welding speed, and wire feed speed of the three welding torches to particle positions in a multi-dimensional search space, and defines the weighted sum of welding quality evaluation indicators and production efficiency indicators as the objective function. Within each preset control cycle, the biomimetic cluster intelligent controller searches for the optimal parameter combination through parallel computation and converts this combination into control commands, which are then sent to the execution layer.

[0011] Preferably, the three independent servo drivers in the execution layer adopt a multi-axis linkage and collaborative control mode, and each servo driver has an independent motion compensation function. When one of the welding torches has a trajectory deviation outside the preset range due to electromagnetic interference or workpiece deformation, the collaborative decision layer calculates the corresponding compensation vector and performs nanosecond-level real-time correction of the welding torch's attitude through the three independent servo drivers to ensure that the synchronization accuracy of the three welding torches is maintained within the preset accuracy range.

[0012] Furthermore, the physical sensing layer is also equipped with an arc feature extraction unit, which is used to analyze the voltage and current waveform characteristics of the welding arc in real time and identify whether there is short-circuit transition or arc blow phenomenon; the arc feature extraction unit feeds back the identification results to the collaborative decision layer, which triggers the spatial intelligent decoupling layer to adjust the gain of the magnetic field strength.

[0013] Furthermore, the collaborative real-time process control system based on three-gun welding equipment also includes a global data storage and analysis terminal, which is used to record all sensing data, control commands and final weld formation quality data during the welding process, and continuously optimize the thermal compensation model and magnetic field neutralization strategy in the collaborative decision-making layer through deep learning of historical data.

[0014] Preferably, each coil unit in the active dynamic magnetic field neutralization module supports independent decoupling control. The collaborative decision layer calculates the composite magnetic field vector at any point in space in real time based on the relative distance and angle between the three welding torches in three-dimensional space, and uses space vector modulation technology to drive each coil unit, thereby achieving precise reconstruction and guidance of the electromagnetic field in a complex three-dimensional working environment.

[0015] Furthermore, the multispectral thermal imaging device in the physical sensing layer has an automatic calibration function, which can automatically adjust the emissivity compensation coefficient for workpieces of different materials and surface roughness to ensure the accuracy of the extracted workpiece surface temperature data. The collaborative decision-making layer uses this accurate temperature data to perform fluid dynamics calculations, predict the wettability and flow direction of the molten pool, and then fine-tune the oscillation amplitude and frequency of each welding torch.

[0016] Furthermore, the collaborative decision-making layer has fault self-diagnosis and safety protection logic. When the environmental parameters monitored by the physical perception layer or the operating status fed back by the execution layer are abnormal and exceed the preset safety boundary, the collaborative decision-making layer immediately issues an emergency shutdown command to the servo driver of the three welding torches, and simultaneously cuts off the welding power supply, and records fault snapshot information on the global data storage and analysis terminal.

[0017] Furthermore, the collaborative real-time process control system based on three-gun welding equipment introduces a timestamp synchronization mechanism to ensure that data exchange between the physical perception layer, spatial intelligent decoupling layer, collaborative decision-making layer, and execution layer follows strict temporal determinism. Within each preset communication cycle, the system completes a complete closed loop of sensing, calculation, decision-making, and execution. The duration of the communication cycle is much shorter than the minimum change cycle of the welding arc physical process, thus achieving true real-time collaborative control.

[0018] Preferably, the collaborative decision-making layer further includes a trajectory planning unit, which is used to automatically calculate the starting point coordinates, ending point coordinates, and intermediate transition trajectories of the three welding torches based on the geometric model of the workpiece to be welded; during the welding operation, the trajectory planning unit dynamically modifies the current motion trajectory based on the weld offset data collected in real time by the physical perception layer, so as to realize real-time automatic tracking of the weld.

[0019] Furthermore, the three independent servo drivers in the execution layer are equipped with high-resolution photoelectric encoders to monitor the actual position of the welding torch tip; the collaborative decision layer compares the actual position with the commanded position in real time and eliminates position tracking errors through a preset proportional-integral-derivative control strategy.

[0020] Furthermore, the spatial intelligent decoupling layer has an adaptive impedance matching function, which can automatically adjust the output characteristics of the compensation current according to the changes in impedance in the welding circuit, prevent the failure of the magnetic field neutralization effect due to drastic fluctuations in the welding load, and ensure that the electromagnetic field neutralization efficiency is always in the preset optimal range throughout the entire welding process.

[0021] Furthermore, the collaborative decision-making layer has multiple pre-stored process expert modes for welding dissimilar metals or complex structural parts with varying thicknesses; the system can identify the characteristics of the current workpiece and automatically call the corresponding collaborative control logic to achieve adaptive process adjustment for different welding objects.

[0022] Preferably, the physical perception layer further includes a sound sensor for collecting the acoustic signals of the electric arc during the welding process. The collaborative decision-making layer analyzes the frequency distribution and energy intensity of the acoustic signals to help determine the stability and penetration state of the electric arc, and integrates the acoustic feedback information into the multi-criteria optimization algorithm of the biomimetic cluster intelligent controller.

[0023] Furthermore, the collaborative real-time process control system based on three-gun welding equipment adopts a distributed control architecture. The physical sensing layer and execution layer are located at the front-end operation site and are interconnected with the collaborative decision-making layer located in the central control cabinet through a high-speed industrial fieldbus. This architecture ensures the communication reliability and data integrity of the system under high-intensity electromagnetic interference environment.

[0024] Furthermore, the collaborative decision-making layer employs multi-field coupling analysis logic when calculating the magnetic field neutralization parameters. It not only considers the static magnetic field generated by the current, but also takes into account the influence of the induced electromotive force generated by the high-speed movement of the electric arc on the magnetic field distribution, thus achieving complete elimination of magnetic field deviation during dynamic welding.

[0025] Furthermore, the execution layer has a flexible expansion interface, which supports increasing or decreasing the number of controlled welding torches according to specific production process requirements; the collaborative decision layer can automatically identify the configuration of the connected devices and reconstruct the corresponding collaborative control topology.

[0026] Preferably, the collaborative decision-making layer sets different collaborative scheduling strategies for long straight welds and curved welds respectively; in the curved welding stage, the system focuses on optimizing the angular velocity matching and centrifugal force compensation of the three welding torches, and dynamically corrects the magnetic field imbalance caused by the change of welding torch angle through the spatial intelligent decoupling layer.

[0027] Furthermore, the components of the collaborative real-time process control system based on the three-gun welding equipment adopt a preset redundancy backup mechanism. When one of the sensors or drive units experiences a non-fatal failure, the collaborative decision-making layer can automatically switch to a safe operation mode and use the remaining sensing resources and actuators to maintain the basic operating functions of the system.

[0028] Compared with the prior art, the present invention has the following beneficial effects: 1. By introducing active dynamic magnetic field neutralization technology of spatial intelligent decoupling layer, this invention eliminates electromagnetic interference during multi-gun synchronous welding, enabling the three welding guns to achieve fully synchronous operation with high time utilization within the time and space window that should interfere with each other; compared with the traditional serial operation or off-peak operation strategy, the production efficiency improvement reaches the predetermined target and shortens the manufacturing cycle of large structural parts.

[0029] 2. Through the deep integration of the physical perception layer and the collaborative decision-making layer, the system achieves real-time and precise control of the electromagnetic field and the thermal field; the active control of the magnetic field distribution solves the problems of weld bead offset, undercut and spatter caused by magnetic blow, while the pre-compensation mechanism based on thermal flow field topology analysis effectively suppresses component deformation caused by thermal stress concentration, reduces thermal deformation, and enables the weld bead formation consistency, penetration depth and mechanical properties of multi-gun synchronous welding to meet the high standard requirements of single operation.

[0030] 3. This invention breaks the long-standing technical prejudice in the field that multi-gun operation inevitably leads to uncontrollable interference. It draws on the field control concept from high-energy physics and meteorology, transforming passive avoidance into active neutralization and guidance, and turning the interfering thermal field into a usable preheating resource, providing a brand-new systematic solution for intelligent manufacturing in multi-physics field coupling environments.

[0031] 4. The system possesses a high degree of intelligence and adaptability: It adopts a biomimetic swarm intelligence algorithm for multi-parameter real-time closed-loop control, enabling the system to cope with complex and ever-changing welding environments and diverse workpiece configurations; through continuous data accumulation and model optimization, the system exhibits strong robustness and process adaptability, reducing reliance on operator experience and improving the automation and intelligence level of the welding process.

[0032] 5. Excellent security and scalability: The distributed control architecture and comprehensive self-diagnosis and redundancy backup mechanisms ensure long-term stable operation of the system in extremely harsh industrial environments; the flexible interface design enables the system to adapt flexibly to the needs of production lines of different sizes, with broad industrial application prospects and significant economic benefits. Attached Figure Description

[0033] Fig. 1 This is a schematic diagram of the overall technical solution architecture according to the present invention; Fig. 2 This is a schematic diagram of the core principle framework of the active dynamic magnetic field neutralization module according to the present invention; Fig. 3 This is a flowchart illustrating the logical flow framework of the collaborative decision-making layer based on the biomimetic cluster optimization strategy according to the present invention. Fig. 4This is a schematic diagram illustrating the multi-level interaction relationship and data flow between the physical perception layer and the collaborative decision-making layer in this invention; Fig. 5 This is a schematic diagram illustrating the principle framework of thermal flow field topology analysis and energy input pre-compensation according to the present invention. Detailed Implementation

[0034] Example 1: To make the objectives, technical solutions, and advantages of the present invention clearer, the following is combined with... Figs. 1 to 5 The present invention will be further described in detail below with reference to specific embodiments.

[0035] The collaborative real-time process control system based on three-gun welding equipment includes a physical sensing layer, a spatial intelligent decoupling layer, a collaborative decision-making layer, and an execution layer. The physical sensing layer is used to capture and digitize multi-dimensional physical field information during the welding process in real time. The spatial intelligent decoupling layer is used to eliminate mutual inductance interference between multiple arcs through active electromagnetic intervention. The collaborative decision-making layer is used to perform optimal scheduling of global parameters based on biomimetic algorithms and thermal flow field analysis. The execution layer is used to drive multiple welding actuators to complete collaborative operations.

[0036] The physical sensing layer specifically includes an electromagnetic field and thermal field fusion sensor array, an arc feature extraction unit, a sound sensor, and a high-speed signal conditioning module. The electromagnetic field and thermal field fusion sensor array further includes multiple high-frequency Hall sensors arranged in a preset array geometric distribution. These Hall sensors are configured around the molten pool of each welding torch and in the gaps between the torches to capture the dynamic synthetic magnetic field generated by the three welding torches operating synchronously. Each Hall sensor has high-frequency response characteristics, and its sampling frequency is set to a value much higher than the welding current ripple frequency, enabling it to record the instantaneous changes in the magnetic induction intensity vector. The multispectral thermal imaging device is installed directly above or diagonally above the welding area and has an automatic calibration function. It can automatically adjust the mapping coefficient between radiation intensity and temperature for workpiece surfaces of different materials such as stainless steel, aluminum alloy, or carbon steel using an internally stored emissivity compensation table. The multispectral thermal imaging device is configured to acquire the spatial distribution of thermal radiation intensity on the workpiece surface and convert it into a dynamic three-dimensional thermal field model with coordinate information. The high-precision displacement monitoring unit uses the principle of laser interference or a high-resolution grating structure to monitor the microscopic deformation of the workpiece during the heating process and the actual height of the welding torch tip relative to the workpiece surface in real time.

[0037] The arc feature extraction unit is integrated within the physical sensing layer, with its input connected to the output of the welding power source. This unit is configured to acquire real-time waveforms of the welding voltage and current at a sampling rate on the order of megahertz. Through a built-in high-speed digital signal processor, the unit can analyze the statistical characteristics of the arc waveform, such as the average voltage, effective current, and crest factor, to identify any abnormal arc oscillations caused by short-circuit transition, jet transition, or magnetic blow. The sound sensor is a high-sensitivity industrial-grade microphone array deployed within the welding noise environment to collect acoustic characteristic signals generated by the arc combustion. The collaborative decision-making layer analyzes the spectral energy distribution of the acoustic signals to assist in determining the stability of the arc.

[0038] The spatial intelligent decoupling layer specifically includes an active dynamic magnetic field neutralization module, a miniaturized Helmholtz coil array, a high-speed, high-current power drive unit, and an adaptive impedance matching unit. The active dynamic magnetic field neutralization module is the core of the system's physical control, configured to change the local electromagnetic environment around each welding torch in real time according to compensation commands issued by the collaborative decision-making layer. The Helmholtz coil array is made of highly conductive oxygen-free copper material and is layered and wrapped around the outside of each welding torch nozzle. The geometric parameters of each coil pair are precisely calculated to ensure a highly uniform controlled magnetic field is generated near the coil's central axis. The high-speed, high-current power drive unit uses a full-bridge converter circuit based on insulated-gate bipolar transistors (IGBTs) or silicon carbide power devices, capable of injecting compensation current with rapidly adjustable phase, amplitude, and frequency into the Helmholtz coils according to the pulse width modulation signal.

[0039] The adaptive impedance matching unit is connected between the power drive unit and the coil array. Due to temperature changes and the movement of the welding torch during welding, the inductive reactance and impedance of the coil circuit will dynamically fluctuate. The adaptive impedance matching unit is configured to monitor the feedback current of the circuit in real time and adjust the matching capacitor or inductor parameters at the output end to ensure that the energy output by the power drive unit can be converted into a compensating magnetic field without loss and quickly. The working principle of the active dynamic magnetic field neutralization module is as follows: it acquires the current vector of two or three adjacent welding torches in real time, calculates the synthetic interference magnetic field in the interference region, and then drives the corresponding coil to generate a canceling magnetic field with opposite direction and equal magnitude. This method does not passively shield the magnetic field, but uses the principle of antiphase superposition to realize the folding and reconstruction of magnetic field lines in physical space, so that the arc of each welding torch is in a quasi-static magnetic field environment that is close to that of a single torch operation.

[0040] The adaptive impedance matching unit acquires the instantaneous voltage signal at the output of the high-speed, high-current power drive unit and the feedback current signal in the Helmholtz coil circuit in real time. The internal processing logic is as follows: The phase difference between the voltage and current fundamental signals is extracted using digital phase-locked loop (PLL) technology, and this phase difference is used as the error input for closed-loop control. The system incorporates an incremental digital proportional-integral (PI) controller, which calculates the digital control word for the adjustable capacitor in the matching impedance network with the phase difference approaching zero. This control word is sent to a solid-state relay array or digital potentiometer via a fieldbus, dynamically changing the capacitance value connected in parallel across the coil array. This counteracts the spatial capacitance changes caused by welding torch movement and the coil inductive reactance drift caused by temperature rise, ensuring that all active power output from the power drive unit is converted into a canceling magnetic field.

[0041] Phase difference extraction and error calculation:

[0042] in, Indicates the first Phase difference error per control cycle This indicates the target phase difference (usually 0, i.e., purely resistive). and These represent the fundamental phasors of the feedback current and voltage, respectively. This indicates the conjugate transpose operation.

[0043] Incremental PI control law (matching capacitor adjustment):

[0044]

[0045] in, Indicates the first Capacitor increment control word per cycle, and These represent the proportional and integral parameters of the digital PI controller, respectively. and These represent the impedance matching capacitance values ​​at the current and previous moments, respectively. Indicates the first Phase difference error per control cycle.

[0046] The collaborative decision-making layer specifically includes a biomimetic cluster intelligent controller, a thermal flow field topology analysis module, a trajectory planning unit, and a fault self-diagnosis and safety protection module. The collaborative decision-making layer adopts a distributed parallel computing architecture, with the core processing unit composed of a high-performance multi-core processor and a field-programmable gate array (FPGA). The thermal flow field topology analysis module utilizes three-dimensional temperature data acquired from the physical sensing layer to construct a fluid dynamics calculation model. This module is configured to identify the location of the heat accumulation center and the evolution trend of the thermal gradient on the workpiece surface. Borrowing from eddy current analysis algorithms in meteorology, it abstracts the temperature field distribution into a combination of scalar and vector fields, calculating the thermal stress coupling region caused by the three-source heat input.

[0047] The thermal flow field topology analysis module receives discrete temperature data points on the workpiece surface uploaded by the physical sensing layer. The system first reconstructs the discrete data points into a continuous three-dimensional temperature scalar field matrix using a three-dimensional interpolation algorithm. Internally, the system calculates the temperature gradient vector field by taking the first-order spatial partial derivative of this scalar field, and then derives the heat flux density vector field by combining this with the thermal conductivity coefficient of the workpiece material. The system accurately locates the coordinates of the "heat accumulation center" where the divergence is greater than zero by calculating the divergence of the heat flux vector field. Subsequently, based on the thermoelastic constitutive equations, the system transforms the temperature gradient field into a virtual thermal stress field. When the calculated local equivalent stress value exceeds the material's yield strength threshold, a pre-compensation logic is triggered: the area integral of the stress-exceeding region is calculated with the heat accumulation center as the center, and this integral value is used as a feedforward compensation coefficient to dynamically generate current reduction step size and speed compensation step size, which are then sent to the biomimetic cluster intelligent controller for secondary parameter optimization.

[0048] Calculation of heat flux density vector field:

[0049] in, Represents the heat flux density vector. Indicates the thermal conductivity of the workpiece material. Indicates temperature. The gradient operator represents the gradient of a three-dimensional temperature scalar field. Represents the spatial position of a point inside an object in a three-dimensional Cartesian coordinate system. Represents time.

[0050] Identification of heat accumulation centers (divergence calculation):

[0051] in, Represents the divergence of the thermal flow field, when The coordinate point is then determined to be the heat accumulation center.

[0052] Calculation of energy input pre-compensation:

[0053]

[0054] in, Indicates the first Current reduction compensation for the welding torch This represents the proportional adjustment coefficient. This indicates that the divergence exceeds the safety threshold. Let $dS$ represent the area of ​​the integral over the region. This indicates that the speed of the remaining welding torches is increased by a compensation amount. This represents the velocity compensation coupling coefficient.

[0055] The biomimetic swarm intelligent controller acts as the brain of the collaborative decision-making layer, incorporating a decision-making strategy based on an improved particle swarm optimization algorithm. In this algorithm, the welding current, welding voltage, moving speed, and wire feed speed of the three welding torches are mapped to the position coordinates of each particle in a high-dimensional search space. Each particle represents a possible parameter combination. The biomimetic swarm intelligent controller is configured to calculate the objective function value under the current parameter combination within each control cycle (e.g., every 10 milliseconds). The objective function is obtained by weighted summation of welding quality indicators (such as the consistency of penetration depth and the smoothness of weld formation) and production efficiency indicators (such as the deposition rate per unit time). Through particle collaboration and information sharing, the algorithm quickly finds the parameter combination that optimizes the objective function. When the thermal flow field topology analysis module predicts that the heat accumulation in a certain area is about to exceed a preset safety threshold, the biomimetic swarm intelligent controller dynamically adjusts the energy input of the corresponding welding torch in that area, such as reducing the welding current, and simultaneously compensates for the moving speed of the other two welding torches to maintain the stability of the overall production cycle.

[0056] The parameter decision module of the biomimetic swarm intelligent controller receives real-time feedback data from the physical perception layer. It combines 12 physical quantities—current, voltage, welding speed, and wire feed speed—from the three welding torches into a 12-dimensional state vector, which serves as the position coordinates of a single particle in the particle swarm optimization algorithm. Internally, the system first performs validity checks on particle positions based on preset process boundary conditions. Then, it inputs the valid particle position vectors into a preset multi-objective fitness evaluation function. This function takes 12-dimensional parameters as input and outputs a single scalar fitness value. To overcome the standard particle swarm optimization algorithm's tendency to get trapped in local optima within a multi-extreme welding parameter space, this invention introduces inertial weights based on a nonlinear decreasing strategy and a Cauchy mutation operator. In each iteration, the particle updates its velocity vector based on its historical best position and the global best position. If the fitness value does not improve for several consecutive generations, a Cauchy mutation is triggered to perturb the current particle's position. After iteration convergence, the optimal particle's position vector is decoded into 12 control pulse commands, which are then transmitted to the three independent servo drivers in the execution layer.

[0057] Target fitness function:

[0058] in, The fitness value represents the particle position vector $X$. Indicates the total number of sampling points. and They represent the first The penetration depth and weld smoothness at each sampling point and This represents the corresponding target ideal value. , , The preset weighting coefficients, Indicates the first The deposition efficiency coefficient of the welding torch. Indicates welding speed. This indicates the total welding time.

[0059] Nonlinear dynamic inertia weight update formula:

[0060] in, Indicates the first Inertia weights in the next iteration and These represent the upper and lower limits of the weight, respectively. This indicates the maximum number of iterations.

[0061] Particle velocity and position update formulas:

[0062]

[0063] and They represent the first The particle in the first The first dimension in space The speed and position of the next iteration and They represent the first The particle in the first The first dimension in space The speed and position of the next iteration and Represents the learning factor. and Represents a random number between [0, 1]. This represents the individual's historical best position. This indicates the globally best historical position. This represents the control coefficient for the probability of variation. Represents a standard Cauchy distribution random number. Indicates the weight.

[0064] The trajectory planning unit is configured to automatically generate the initial motion paths of the three welding torches based on the computer-aided design (CAD) model of the workpiece. During welding, the unit receives weld position deviation data from the physical sensing layer and uses an inverse kinematics algorithm to calculate the real-time compensation required for the servo axes, achieving automatic tracking. The fault self-diagnosis and safety protection module is configured to monitor the system's operating status throughout the process. If current fluctuations exceed the rated range, coil temperature is too high, or communication network latency is abnormal, the module will immediately trigger emergency shutdown logic, cutting off the welding power supply and locking the servo motors to ensure equipment and personnel safety.

[0065] The execution layer specifically includes three independent servo drivers for the welding torches, a multi-axis motion mechanism, a welding execution mechanism, and a high-resolution photoelectric encoder. The three independent servo drivers possess high dynamic response characteristics and support multi-axis linkage and collaborative control modes. Each servo driver is responsible for controlling the six-degree-of-freedom or three-degree-of-freedom motion of one welding torch. The high-resolution photoelectric encoder is mounted at the end of the servo motor to provide real-time feedback of the welding torch's precise spatial coordinates. The collaborative decision layer compares the commanded position with the actual position fed back by the encoder in real time, using a proportional-integral-derivative (PID) control strategy to eliminate lag and errors generated during mechanical transmission. The welding execution mechanism includes the welding torch itself and a matching wire feeding mechanism. The wire feeding mechanism is controlled by the execution layer and can achieve pulsed or constant-speed wire feeding according to the instructions of the decision layer to coordinate with changes in current.

[0066] Furthermore, the collaborative real-time process control system based on the three-gun welding equipment achieves data synchronization between different levels by introducing a unified time reference signal. The physical sensing layer adds nanosecond-level timestamps to the data acquisition. Upon receiving this data, the collaborative decision-making layer aligns the data based on the timestamps, ensuring that the information upon which the magnetic field compensation algorithm and the thermal compensation logic are based is completely matched in the time domain. The system also includes a global data storage and analysis terminal, which utilizes high-capacity solid-state storage to record all raw sensing data, control commands, and post-weld quality inspection data during the welding process. This global data storage and analysis terminal is configured to run deep learning algorithms, continuously optimizing the objective function model and preset parameters of the magnetic field neutralization strategy in the biomimetic swarm intelligent controller through offline training on massive amounts of historical data.

[0067] In actual operation scenarios, when three welding torches are activated simultaneously, the high-frequency Hall sensor array in the physical perception layer immediately captures the coupled magnetic fields generated by the three strong currents. Based on this real-time current vector information, the spatial intelligent decoupling layer quickly calculates the coil current parameters required to counteract these interfering magnetic fields. Simultaneously, the thermal field data of the workpiece surface captured by the thermal imaging device is transmitted to the collaborative decision-making layer. The thermal flow field topology analysis module calculates that if the three torches operate with preset parameters, the central region of the workpiece will experience a risk of deformation due to thermal stress concentration. At this point, the biomimetic cluster intelligent controller immediately performs optimization calculations, reducing the current intensity of the middle welding torch and appropriately increasing its oscillation amplitude to disperse heat, while fine-tuning the welding voltage of the two side torches to compensate for the loss of deposition. Upon receiving these complex instructions, the execution layer precisely controls the attitude and electrical parameters of the welding torches through servo drives. The entire process completes a closed loop within milliseconds.

[0068] Example 2: As a further supplement and evolution of Example 1, Example 2 describes a collaborative real-time process control system based on a three-gun welding equipment using a distributed edge computing architecture. In this example, the system's control logic no longer relies entirely on the central collaborative decision-making layer, but instead pushes some of the sensing and decoupled control functions down to the edge processing units corresponding to each welding gun, thereby further reducing the system's communication latency and improving real-time performance.

[0069] In Embodiment 2, the system includes a distributed physical sensing network, an edge intelligent decoupling module, a global collaborative management layer, and a distributed execution unit. The distributed physical sensing network integrates an independent edge processing node on the end effector of each welding torch. This node is directly connected to the Hall sensor, current transformer, and near-field infrared temperature measurement unit of that welding torch. The edge processing node is configured to autonomously process the local physical quantity characteristics of its welding torch, eliminating the need to upload all raw high-frequency data to a central server, thus significantly alleviating the bandwidth pressure on the fieldbus.

[0070] The edge intelligent decoupling module is integrated into the drive control box of each welding torch. Internally, it contains a dedicated application-specific integrated circuit (ASIC) or high-performance digital signal processor (DSP) for magnetic field calculation. This edge intelligent decoupling module is configured to exchange current parameters in real time with the edge modules of adjacent welding torches via a high-speed peer-to-peer communication link. Based on the exchanged adjacent current information, the module directly calculates the required compensation for the Helmholtz coil locally and directly drives the power unit output. This distributed architecture improves the response frequency of magnetic field neutralization, enabling it to cope with transient magnetic field interference caused by high-frequency pulse currents in the welding arc.

[0071] In this embodiment, the global collaborative management layer is primarily responsible for high-level logic scheduling and thermal field balance control. It interconnects with each edge processing node via industrial Ethernet to acquire compressed thermal field characteristic data and the operational progress of each welding torch. The global collaborative management layer incorporates a path coordination strategy based on an ant colony algorithm. During the welding of complex curved seams, this strategy treats each welding torch as an individual ant with autonomous decision-making capabilities. By simulating the pheromone release and perception mechanisms of ants searching for paths, it coordinates the relative positions of the three welding torches in space, avoiding mechanical interference while ensuring that the heat input distribution on the workpiece surface reaches global optimum.

[0072] The global collaborative management layer also features a self-organizing reconfiguration function. When an edge processing node of a welding torch in the system detects an unrecoverable hardware failure, the node will send an alarm to the global collaborative management layer. The management layer will then initiate a reconfiguration logic, automatically replanning the work paths of the other two welding torches and adjusting the welding speed and current intensity to take over the unfinished tasks of the faulty welding torch, ensuring the continuity of the production line.

[0073] Each servo driver in the distributed execution unit is equipped with a controller featuring a real-time Ethernet interface. These controllers support a distributed clock synchronization protocol, ensuring that synchronization errors in multi-axis motion are suppressed to the microsecond level. Simultaneously, each execution unit has a local safety protection loop, enabling it to enter a safe standby mode based on feedback from local sensors in the event of a loss of global command signals.

[0074] In Example 2, the neutralization process of the electromagnetic field is described in more detail. The edge intelligent decoupling module employs spatial vector modulation technology in its calculation of the compensation magnetic field. Specifically, the system divides the three-dimensional spatial coordinate system into multiple sectors and calculates the conduction time ratio of each coil unit in adjacent coil arrays based on the projection position of the interference magnetic field vector in the coordinate system. In this way, a rotating magnetic field of arbitrary direction and intensity can be generated around the welding torch, achieving precise purification of the constantly changing electromagnetic environment under complex welding trajectories.

[0075] In the edge intelligent decoupling module of Embodiment 2, after receiving the three-dimensional cancellation magnetic field vector command issued by the collaborative decision-making layer, the system first projects the magnetic field target vector in three-dimensional space onto the two-phase stationary coordinate system through the inverse Clarke transformation. In a coordinate system, two orthogonal magnetic field components are obtained. The internal algorithm logic will... The plane is divided into six sectors. The sector number of the current target vector is determined by judging the polarity and algebraic relationship between the two components. Then, based on the sector, the duration ratio of the interaction between the two adjacent fundamental non-zero magnetic field vectors and the zero vector is calculated using the volt-second balance principle. Finally, these durations are converted into pulse-width modulation duty cycle signals for the corresponding micro-coil units and output to the power drive unit, thereby maximizing the utilization of the coil's magnetic induction intensity output capability without changing the DC bus voltage.

[0076] Projection of two-phase stationary coordinate systems:

[0077] in, This indicates that the three-phase expectation issued by the collaborative decision-making level will cancel out the magnetic field component. This represents the two-phase orthogonal magnetic field components after transformation.

[0078] Sector determination logic:

[0079] in, Indicates the sector number (values ​​1-6). It is a symbolic function.

[0080] Calculation of the interaction time of adjacent vectors (taking the first sector as an example):

[0081]

[0082]

[0083] in, , These represent the durations of action of two adjacent fundamental non-zero voltage vectors, respectively. Indicates the time of action of the zero vector. Indicates the PWM control period. This represents the DC bus voltage of the power drive unit.

[0084] The pre-compensation logic for the thermal flow field is also enhanced in Example 2. The global collaborative management layer is configured to establish a nonlinear mapping model between workpiece material, plate thickness, ambient temperature, and thermal strain through in-depth analysis of historical thermal field data. This model is embedded in the inference engine of the management layer. Before welding starts, the system pre-calculates a thermal balance reference based on the input process parameters. During welding, the system only needs to monitor the deviation between the actual thermal field and the reference thermal field, and uses predictive control strategies to proactively correct the deviation, keeping the thermal deformation of the workpiece within a preset small range.

[0085] Example 3: Example 3 details a collaborative real-time process control system for a three-gun welding equipment based on digital twin-driven architecture. In this example, the system introduces a high-precision virtual simulation model and a real-time bidirectional mapping of real-time data. By performing advanced simulation in the virtual space, it guides the actual welding process in the physical space.

[0086] In Embodiment 3, the collaborative decision-making layer further includes a digital twin engine module. This engine module stores a geometric model, kinematic model, and multiphysics coupling numerical model that are completely consistent with the physical device. The digital twin engine module is configured to receive a real-time full data stream from the physical perception layer and reconstruct the current welding state in real time in virtual space, including the shape of each arc, the flow characteristics of the molten pool, and the residual stress distribution inside the workpiece.

[0087] The digital twin engine module possesses advanced predictive capabilities. While the physical equipment is running, this module utilizes its powerful parallel computing capabilities to rapidly simulate the welding process over a future period in virtual space. If the simulation results indicate a risk of weld undercut or internal porosity at some future time under the existing control parameters, the system immediately attempts various optimized parameter combinations in the virtual space. Once the optimal solution that mitigates the quality risk is found, this parameter combination is immediately sent as a control command to the execution layer.

[0088] In Embodiment 3, the physical sensing layer adds a three-dimensional deformation detection unit based on structured light scanning. This unit is configured to perform real-time scanning of the welded bead and the area to be welded during the welding gap or welding process, generating high-precision point cloud data. The collaborative decision-making layer compares the point cloud data with the original design model to calculate the real-time shrinkage and torsion of the workpiece due to heat input. Based on this real-time geometric deformation data, the trajectory planning unit dynamically corrects the travel reference plane of the three welding torches to ensure that the arc is always at the center of the weld.

[0089] In Embodiment 3, the spatial intelligent decoupling layer employs an auxiliary magnetic guiding structure constructed from superconducting magnet technology or high-permeability materials to enhance the effectiveness of magnetic field neutralization. The power drive unit features a higher-dimensional modulation algorithm, capable of not only canceling the magnetic field generated by the steady-state current but also suppressing electromagnetic pulse noise caused by the high-frequency switching of the welding power supply. The adaptive impedance matching unit utilizes a model reference adaptive control algorithm, capable of adjusting the output waveform of the compensation current within nanoseconds based on the inductance and resistance parameters identified in real-time within the circuit.

[0090] The collaborative decision-making layer also integrates an augmented reality (AR)-based human-computer interaction interface. Operators can wear AR glasses to view in real time the 3D thermal field distribution map, electromagnetic force line distribution map, and the system's preset ideal welding path hidden behind the strong arc light. This visualization method makes the debugging process more intuitive. Simultaneously, the system has a knowledge graph-based expert support system. When encountering rare welding defects or complex process problems, the system automatically retrieves historical cases and expert rules from the knowledge graph, providing heuristic control strategies for the collaborative decision-making layer.

[0091] The servo motors in the execution layer are intelligent integrated motors with integrated drive functions. These motors integrate control logic for position, speed, and current loops and possess self-tuning capabilities. The collaborative decision-making layer only needs to issue high-level motion trajectory descriptions; the underlying motion control details are autonomously handled by the intelligent motors, improving the system's modularity and maintenance convenience.

[0092] In the logic of Example 3, to address the complexity of three-gun coordination, the system also designs a conflict coordination mechanism based on game theory. When the three welding torches are operating in a confined space and their respective quality objectives conflict (for example, welding torch 1 needs to increase the current to ensure penetration, but this will lead to excessive thermal stress at the location of welding torch 2), the coordination decision layer is configured to use the Nash equilibrium algorithm to calculate the optimal balance point of compromise, so that the overall operating effect of the three welding torches reaches the expected performance envelope.

[0093] Example 4: Example 4 details a collaborative real-time process control system based on a three-gun welding equipment, possessing self-healing and process self-evolution capabilities. This example focuses on the system's stability during long-term operation and its adaptive learning capability for new materials and processes.

[0094] In Example 4, the physical sensing layer further integrates an online environmental monitoring module. This module is used to monitor humidity, air pressure, and ambient magnetic field background noise within the plant. The collaborative decision-making layer is configured to incorporate environmental factors into the boundary conditions of the control strategy. For example, in a high-humidity environment, the system automatically fine-tunes the arc voltage fluctuation threshold to compensate for minor shifts in arc ionization characteristics.

[0095] In Embodiment 4, the spatial intelligent decoupling layer employs a segmented coil structure. The coil array around each welding torch is divided into multiple independently controllable micro-coil units. The collaborative decision-making layer, based on the welding torch's tilt angle and the relative positions of adjacent welding torches, precisely calculates the magnetic field strength that each micro-coil unit in space should contribute using a spatial vector projection algorithm. This refined control method enables complete neutralization of non-uniform interference magnetic fields during the extremely complex three-dimensional curved surface welding process.

[0096] The collaborative decision-making layer incorporates a process evolution model based on online reinforcement learning. After each welding task, the system obtains an internal quality score for the weld seam using automated inspection equipment (such as ultrasonic testing or online X-ray testing systems) in the physical perception layer. The process evolution model is configured to use these quality scores as reward signals to adjust the internal parameter weights of the biomimetic swarm intelligent controller. As the welding workload increases, the system can automatically learn the best practices for specific materials and structures, achieving a fundamental shift from human experience-driven to data-driven intelligence.

[0097] The process evolution model in Example 4 is constructed as a Markov decision process. The state space is defined as a dimensionality-reduced vector of multidimensional sensor data within the current control cycle (including the mean temperature gradient, magnetic field deviation rate, and arc acoustic energy); the action space is defined as a set of discrete fine-tuning step sizes applied to the current and voltage parameters; the reward function is defined as the weighted difference between the post-weld non-destructive testing score and the current energy consumption index. During internal flow, the Actor policy network receives the state vector and outputs the probability distribution of each action. The system samples and executes an action according to probability. After the environmental (i.e., welding process) state transitions, the Critic value network evaluates the value of the state-action pair. The system uses temporal difference (TD) error to update the value parameters of the Critic network and the policy gradient of the Actor network, enabling the model to spontaneously output the process parameter combination that maximizes the reward when facing similar thermal field distributions in the future.

[0098] Reward function construction:

[0099] in, Indicates time Execute action The instant reward obtained afterward This indicates the non-destructive testing quality score (0-100 points) for the corresponding weld. This indicates the total power consumption of the three welding torches during this control cycle. and This represents the weighting coefficients for quality priority and energy consumption penalty.

[0100] Temporal difference error and policy gradient update:

[0101]

[0102]

[0103] in, Indicates timing difference error. Indicates the discount factor. and This represents the Critic value network's assessment of the current state and the next state. and These represent the parameters of the Critic network and the Actor network, respectively. and This represents the corresponding learning rate. Indicates the Actor policy network in state Select action The probability of.

[0104] The execution layer in Embodiment 4 features a flexible robotic arm linkage function. In addition to the standard three-way welding torch, the system can also support the synchronous operation of auxiliary root cleaning mechanisms or post-weld inspection mechanisms. The collaborative decision-making layer is configured to dynamically allocate computing resources, integrating these auxiliary mechanisms into a unified collaborative control bus. Each driver in the execution layer has an energy feedback function, converting excess kinetic energy into electrical energy during braking and storing it in the system's intermediate DC bus, thus improving the overall energy efficiency ratio of the system.

[0105] In Example 4, the fault self-diagnosis and safety protection module is upgraded to a predictive maintenance system with health management functions. This system monitors the vibration spectrum of the motor bearings, the insulation resistance of the coils, and the heat generated by the controller, using a Long Short-Term Memory (LSTM) neural network to predict the remaining service life of each key component. When a component is predicted to fail in the near future, the system automatically generates maintenance suggestions on the global data storage terminal and orders spare parts in advance.

[0106] In Example 4, the fault self-diagnosis and health management module collects the vibration acceleration time-domain signal of the servo motor bearing. After Fast Fourier Transform (FFT), the energy values ​​within a specific frequency band are extracted as features to construct a sliding time window feature sequence. This feature sequence is input as an input tensor into the LSTM network. Inside the network structure, the data flows sequentially through the forget gate, input gate, and output gate. The forget gate determines which historical degradation features in the cell state at the previous time step need to be discarded; the input gate determines which new degradation signs in the vibration features at the current time step need to be written into the cell state; the output gate determines the hidden layer features output at the current time step based on the updated cell state, and these hidden layer features characterize the current degradation potential of the device. The last layer of the LSTM network is connected to a fully connected layer, mapping the hidden layer features to a health factor scalar value between 0 and 1. During training, labeled historical failure data (with a health factor of 0 when failure occurs) is used, and the mean squared error loss function is employed to update the network weights using the backpropagation time time (BPTT) algorithm.

[0107] Forward propagation gating mechanism:

[0108]

[0109]

[0110]

[0111]

[0112]

[0113] in, Indicates the first The vibration frequency domain energy feature vector input at each time step (dimension: ...) ), and These represent the hidden layer state vectors from the previous and current time steps, respectively (with dimensions of ). ), and Let represent the cell state vectors at the previous and current time steps, respectively. , , Let represent the activation vectors of the forget gate, input gate, and output gate, respectively. Indicates the state of candidate cells. , , , This represents the corresponding weight matrix. , , , This represents the bias vector. This represents the Sigmoid activation function. This represents the hyperbolic tangent activation function. This represents the Hadama product (element-level multiplication).

[0114] Training loss function:

[0115] in, This represents the mean squared error loss value. This represents the total number of training samples. This represents the predicted health factors output by the network. This indicates the true health factor of the label (0 indicates invalid, 1 indicates brand new).

[0116] The system communication network in Example 4 employs industrial-grade redundant Ethernet with Time-Sensitive Networking (TSN) features. By incorporating strict prioritization and scheduling logic into data frames, the system ensures absolutely deterministic latency for critical traffic such as control commands. Even with a large amount of high-definition video data transmission in the network, the synchronization accuracy of the three-gun coordination remains unaffected.

[0117] To address the complexity of multi-physics coupling, the collaborative decision-making layer in Example 4 employs a multi-level decoupling strategy. The first layer is real-time hard decoupling of the electromagnetic field, achieved through a spatial intelligent decoupling layer; the second layer is temporal soft decoupling of the thermal field, achieved through the energy prediction algorithm of the collaborative decision-making layer; and the third layer is spatial compensation decoupling for mechanical deformation, achieved through real-time trajectory correction of the execution layer. These three decoupling mechanisms work together to form a complete, multi-dimensional control loop.

[0118] When welding dissimilar metals, such as joining aluminum alloys to steel plates, the system in Example 4 can automatically invoke a pre-stored dissimilar metal expert mode. In this mode, the polarity, pulse frequency, and magnetic field neutralization strategy of the three welding torches are asymmetrically adjusted according to the differences in the thermophysical properties of the materials on both sides. The physical sensing layer focuses on monitoring the formation of intermetallic compounds in the interface region and controls the heat input by adjusting the laser heat source or arc parameters to prevent the formation of brittle phases.

[0119] The above specific embodiments are merely illustrative examples of the technical solutions of the present invention. It should be emphasized that the connection relationships between the various modules, the internal logic algorithms, and the hardware selection of the collaborative real-time process control system based on three-gun welding equipment involved in the present invention can all be equivalently replaced and combined according to the actual industrial application requirements without departing from the core concept of the present invention.

Claims

1. A collaborative real-time process control system based on three-gun welding equipment, characterized in that, It includes a physical perception layer, a spatial intelligent decoupling layer, a collaborative decision-making layer, and an execution layer. The physical perception layer includes an electromagnetic field and thermal field fusion sensor array, which is used to monitor in real time the dynamic welding current vector, spatial pose coordinates, local magnetic induction intensity distribution of the welding area, and real-time three-dimensional temperature gradient of the workpiece surface generated by the three welding torches, and to convert the collected physical quantity information into standardized digital signals and transmit them to the collaborative decision-making layer in real time. The spatial intelligent decoupling layer includes an active dynamic magnetic field neutralization module, which is distributed in the outer periphery of each welding torch. It generates an anti-phase electromagnetic field according to the control command output by the collaborative decision layer to cancel the mutual inductance magnetic field interference between adjacent welding torches, thereby achieving independent purification of the welding arc environment in the spatial dimension. The collaborative decision-making layer includes a biomimetic cluster intelligent controller, which receives multi-dimensional sensor data fed back by the physical perception layer. Based on the thermal flow field topology analysis algorithm and the biomimetic cluster optimization strategy, it calculates in real time the optimal electrical parameter combination of each welding torch, the spatial path compensation amount, and the magnetic field neutralization parameter of the spatial intelligent decoupling layer, thereby realizing nonlinear closed-loop control of the entire welding process. The execution layer includes three independent servo drivers for welding torches and corresponding welding execution mechanisms. The three independent servo drivers are used to drive the three welding torches to perform synchronous welding operations according to the instructions issued by the collaborative decision-making layer, and to provide real-time feedback on the motion status of each drive axis to the collaborative decision-making layer.

2. The collaborative real-time process control system based on three-gun welding equipment according to claim 1, characterized in that: The electromagnetic field and thermal field fusion sensor array includes multiple high-frequency Hall sensors, a multispectral thermal imaging device, and a high-precision displacement monitoring unit. The high-frequency Hall sensor is installed in a preset array distribution around the molten pool of each welding torch and in the gap between the welding torches. Its sampling frequency is higher than the ripple frequency of the welding current, and it is used to capture the transient magnetic field changes caused by the welding current fluctuation when the three welding torches are operating synchronously in real time. The multispectral thermal imaging device is installed above the welding area and is equipped with an automatic calibration module, which is used to automatically adjust the emissivity compensation coefficient for workpiece surfaces of different materials, obtain the thermal radiation intensity distribution of the workpiece surface, and construct a dynamic three-dimensional thermal field model. The high-precision displacement monitoring unit uses a laser interference component to monitor the microscopic deformation of the workpiece during the welding heating process and the actual height change of the welding torch tip relative to the workpiece surface in real time.

3. The collaborative real-time process control system based on three-gun welding equipment according to claim 2, characterized in that: The active dynamic magnetic field neutralization module includes a miniaturized Helmholtz coil array, a high-speed, high-current power drive unit, and an adaptive impedance matching unit. The Helmholtz coil array is made of oxygen-free copper material with high conductivity and is wrapped in layers around the nozzles of each welding gun. The high-speed, high-current power drive unit employs a full-bridge converter circuit based on insulated-gate bipolar transistors or silicon carbide power devices. It is connected to the collaborative decision layer and is used to inject a compensation current with the opposite phase and matching amplitude to the interference magnetic field into the Helmholtz coil array according to the pulse width modulation signal, thereby generating a locally controlled canceling magnetic field. The adaptive impedance matching unit is connected between the power drive unit and the coil array. It is used to monitor the feedback current of the coil circuit in real time and dynamically adjust the circuit matching parameters at the output end to compensate for the fluctuation of the circuit inductive reactance caused by temperature changes or welding torch movement.

4. The collaborative real-time process control system based on three-gun welding equipment according to claim 3, characterized in that: The collaborative decision-making layer has a built-in thermal flow field topology analysis module, which is used to execute pre-compensation logic based on thermal flow field topology analysis. The pre-compensation logic analyzes the real-time temperature distribution data uploaded by the physical sensing layer, abstracts the temperature field distribution into a combination of scalar and vector fields, identifies the heat accumulation center and heat flow field evolution trend on the workpiece surface, and dynamically calculates the instantaneous energy input adjustment of each welding torch accordingly. When the predicted thermal stress in a region exceeds a preset safety threshold, the collaborative decision-making layer issues an instruction to reduce the welding current of the corresponding welding torch in that region and simultaneously increases the travel speed of the other two welding torches to maintain the dynamic balance of the thermal field on the overall welding trajectory.

5. The collaborative real-time process control system based on three-gun welding equipment according to claim 4, characterized in that: The biomimetic cluster intelligent controller has a built-in parameter decision module based on an improved particle swarm optimization algorithm. The parameter decision module maps the current, voltage, welding speed, and wire feed speed of the three welding guns to particle position coordinates in a multi-dimensional search space, and defines the weighted sum of the welding quality evaluation index and the production efficiency index as the objective function. Within each preset control cycle, the biomimetic cluster intelligent controller searches for the parameter combination that makes the objective function reach its optimal value through parallel computing, and converts the parameter combination into specific control pulse commands and sends them to the execution layer. The welding quality evaluation indicators include the consistency parameter of penetration depth and the flatness parameter of weld formation.

6. The collaborative real-time process control system based on three-gun welding equipment according to claim 5, characterized in that: The physical sensing layer also includes an arc feature extraction unit and a sound sensor; The input of the arc feature extraction unit is connected to the output of the welding power source. It is used to acquire the real-time waveforms of welding voltage and welding current at a sampling rate of megahertz. The built-in digital signal processor analyzes the average voltage, effective current, and peak factor to identify whether there is short-circuit transition or arc blow. The identification results are fed back to the collaborative decision layer in real time, which triggers the spatial intelligent decoupling layer to dynamically adjust the gain of the magnetic field strength. The sound sensor employs an industrial-grade microphone array deployed within the welding operation environment to collect acoustic signals generated by the electric arc combustion. The collaborative decision-making layer analyzes the frequency distribution and energy intensity of the acoustic signals to assist in judging the stability and penetration state of the electric arc.

7. The collaborative real-time process control system based on three-gun welding equipment according to claim 6, characterized in that: The three independent servo drives in the execution layer adopt a multi-axis linkage and cooperative control mode, and each servo drive is equipped with a high-resolution photoelectric encoder. The photoelectric encoder is installed at the end of the servo motor and is used to monitor the real-time precise spatial coordinates of the welding torch end. The collaborative decision-making layer compares the actual position fed back by the photoelectric encoder with the command position in real time, and eliminates the position tracking error in the mechanical transmission process through a preset proportional-integral-derivative control strategy; When one of the welding torches deviates from its trajectory outside the preset range, the collaborative decision layer calculates the corresponding motion compensation vector and drives the three independent servo drivers to correct the position of the welding torch in real time, ensuring that the synchronization accuracy of the three welding torches is maintained within the preset micron-level accuracy range.

8. The collaborative real-time process control system based on three-gun welding equipment according to claim 7, characterized in that: The system also includes a global data storage and analysis terminal and a timestamp synchronization mechanism; The global data storage and analysis terminal is connected to the collaborative decision-making layer via a high-speed industrial fieldbus. It is used to record full sensing data, control commands, and post-weld inspection quality data during the welding process using a large-capacity storage medium. It also continuously optimizes the thermal compensation model parameters in the collaborative decision-making layer through offline deep learning training on historical data. The timestamp synchronization mechanism is used to add a nanosecond-level time tag to the moment when the physical sensing layer collects data. The collaborative decision-making layer aligns the data according to the time tag to ensure that the sensing, computing, decision-making and execution links follow strict time determinism in the time domain, and the duration of the communication cycle is less than the minimum change cycle of the welding arc physical process.

9. The collaborative real-time process control system based on three-gun welding equipment according to claim 8, characterized in that: The collaborative decision-making layer also includes a digital twin engine module and a trajectory planning unit; The digital twin engine module stores geometric models, kinematic models, and multi-physics coupling numerical models consistent with the physical device. It is used to receive the full data stream of the physical perception layer and reconstruct the welding state in real time in the virtual space. It performs advanced rolling simulation through parallel computing to predict welding quality risks at future moments and generate optimization parameters in advance. The trajectory planning unit is used to automatically calculate the initial motion path of the three welding torches based on the geometric model of the workpiece to be welded, and combined with the point cloud data obtained by the three-dimensional deformation detection unit based on structured light scanning in the physical perception layer, dynamically correct the walking reference plane of the three welding torches to achieve real-time spatial compensation for the thermal deformation of the workpiece.

10. The collaborative real-time process control system based on three-gun welding equipment according to claim 9, characterized in that: The spatial intelligent decoupling layer adopts a segmented coil structure, and the coil array around each welding gun is divided into multiple independent and controlled micro coil units. The collaborative decision-making layer calculates the composite magnetic field vector at any position in space based on the relative distance, tilt angle, and included angle of the three welding torches in three-dimensional space using a spatial vector projection algorithm, and independently drives each micro coil unit using spatial vector modulation technology to achieve accurate reconstruction of non-uniform interference magnetic field. The collaborative decision-making layer also integrates a fault self-diagnosis and safety protection module. It uses a long short-term memory neural network model to monitor the vibration spectrum of the motor bearing and the insulation resistance of the coil in the execution layer, predict the remaining service life of key components, and trigger emergency shutdown logic when the operating status exceeds the safety boundary, cut off the welding power supply and lock each drive shaft.

Images