Aluminum alloy pipe fitting quick connection welding control method and system

Abstract

The invention relates to the field of mechanical engineering, discloses a quick-connection welding control method and system for aluminum alloy pipe fittings, and aims to solve the problems that in the prior art, heat input control is not accurate, the positioning and centering efficiency is low, parameter adjustment lacks self-adaptability, and the multivariable cooperative control capacity is insufficient. The method comprises the following steps: acquiring point cloud data of the end surface of a pipe fitting through three-dimensional scanning, and calculating a six-degree-of-freedom butt joint deviation by adopting an improved iterative nearest point algorithm; a compensation track is generated based on multi-objective optimization, and the six-degree-of-freedom parallel robot is driven to achieve fine adjustment; and after the butt joint deviation is stable, welding is started according to a sectional type current increasing strategy, and the wire feeding and welding speed is dynamically adjusted in combination with fusion depth feedback and visual detection. According to the scheme, high-precision automatic butt joint, stable molten pool control and online defect closed-loop treatment are achieved, the welding quality consistency and the production efficiency are remarkably improved, and the method is suitable for efficient and high-quality connection of various aluminum alloy pipe fittings.

Description

Technical Field

[0001] This invention belongs to the field of mechanical engineering, specifically relating to a quick-connect welding control method and system for aluminum alloy pipe fittings. Background Technology

[0002] With the continuous development of modern manufacturing, lightweight structural design has been widely used in aerospace, rail transportation and new energy vehicles. Aluminum alloys, due to their low density, high specific strength and excellent corrosion resistance, have become one of the preferred materials for key structural components. Among them, aluminum alloy pipe fittings, as core components that carry functions such as fluid transmission and structural connection, directly affect the safety and durability of the overall system due to their connection reliability.

[0003] Welding technology for aluminum alloy pipe fittings is a key process in manufacturing, aiming to achieve high-strength, high-sealing, and deformation-controllable joint quality. However, traditional welding methods face significant challenges in dealing with thin-walled pipe fittings, complex spatial layouts, and mass production requirements. Existing technologies mainly suffer from the following problems: First, the heat input during welding is difficult to control precisely, easily leading to defects such as uneven penetration, burn-through, or lack of fusion, especially at diameter changes or irregularly shaped interfaces. Second, weld positioning and fixture alignment rely on manual adjustment, resulting in low repeatability and difficulty in meeting the requirements of automated production lines for rapid connection. Third, welding parameters are mostly based on experience, lacking a real-time feedback adjustment mechanism for material condition, ambient temperature and humidity, and arc dynamics, leading to poor process stability. Finally, the lack of a coordinated control strategy for welding path, current and voltage sequence, and shielding gas flow during rapid connection results in large fluctuations in the mechanical properties of the joint and difficulty in ensuring consistency. These problems seriously restrict the technological progress of efficient and high-quality connection of aluminum alloy pipe fittings.

[0004] To address the problems existing in current aluminum alloy pipe welding technology, such as inaccurate heat input control, low positioning and alignment efficiency, lack of adaptive parameter adjustment, and insufficient multi-variable collaborative control capability, it is urgent to develop a quick-connect welding control method and system for aluminum alloy pipes, which has significant technical demand and important engineering application value. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of existing technologies and provide a quick-joint welding control method and system for aluminum alloy pipe fittings, which can effectively solve the problems in the background technology. To achieve the above objective, this invention provides the following technical solution: On one hand, a quick-joint welding control system for aluminum alloy pipe fittings, comprising the following components: a real-time data acquisition module, used to acquire spatial pose data and contact edge information of the aluminum alloy pipe fittings during assembly; a geometric matching analysis module, connected to the real-time data acquisition module, used to evaluate the alignment of the geometric features of the end faces of the two pipe fittings to be welded based on a point cloud registration algorithm, generating an initial docking deviation matrix; a dynamic path planning module, receiving the initial docking deviation matrix, combining it with a preset welding process parameter library, calculating a compensatory adjustment trajectory, and outputting execution instructions; an automatic adjustment execution mechanism, responding to the output instructions of the dynamic path planning module, driving the fixture to perform real-time fine-tuning of the spatial position and orientation of the pipe fittings until it reaches the allowable docking tolerance range; and a welding start judgment module. The system monitors the docking status signal. When the docking deviation remains stable within a threshold for an extended period exceeding a preset time window, the welding equipment is triggered to enter a ready state. The welding process control module, upon receiving the ready confirmation signal, starts the welding power supply according to a segmented current increment strategy, while simultaneously monitoring arc stability and penetration feedback signals, dynamically adjusting wire feed speed and welding speed to maintain dynamic balance in the molten pool. The online quality assessment module collects thermal imaging sequences and acoustic emission signals of the weld area after welding, using a pattern recognition model to determine the presence of defects such as incomplete fusion, porosity, or cracks. The defect response processing module, connected to the online quality assessment module, automatically generates a rework path and schedules a repair welding unit to perform local repair operations when defects exceeding the standard are identified. All modules interact via industrial real-time Ethernet to ensure deterministic transmission and synchronous execution of control commands. Preferably, the real-time data acquisition module adopts a multi-view structured light three-dimensional scanning sensor array, which is arranged on the circumferential fixed bracket of the assembly station. The scanning range covers the extension area of ​​the pipe end from 0 to 150 mm, the spatial sampling resolution is set to 0.1 mm, the data refresh frequency is 200 Hz, and the acquired data is converted into an ordered point cloud set in a unified coordinate system after denoising. Furthermore, the point cloud registration algorithm executed in the geometric matching analysis module adopts an improved iterative nearest point algorithm. Its optimization objective function introduces a normal vector consistency weight term and a curvature change constraint term. The iteration termination condition is that the mean square error is less than 0.02 square millimeters or the change in three consecutive iterations is less than 0.005 mm. The resulting initial docking deviation matrix contains 6 degrees of freedom of displacement and rotation components, with an accuracy of 0.01 mm and 0.05°. In addition, the dynamic path planning module has a built-in multi-objective optimization solver. With the optimization objectives of minimizing adjustment energy consumption, shortest adjustment time and maximum fixture load balance, it generates a nonlinear compensation trajectory. This trajectory is decomposed into 5 Bézier curve smooth transition paths. The interpolation period of each path is 2 ms. The path planning result includes a four-dimensional control sequence of time-displacement-velocity-acceleration. Preferably, the automatic adjustment actuator is composed of a 6-DOF parallel robot, with its end flange connected to a flexible clamping unit. The clamping force control range is 200 N to 800 N, the position repeatability accuracy is ±0.008 mm, the motion response delay is less than 10 ms, and the actual pose data is fed back to the dynamic path planning module in real time during the execution process to form a closed-loop correction. Furthermore, the welding start determination module is equipped with a dual-condition triggering mechanism, which requires that the spatial docking deviation be stable for more than 300 ms within the range of ±0.1 mm and ±0.2°, and that the clamp pressure fluctuation amplitude be less than 5% of the rated value before the welding enable signal can be output, so as to avoid false start due to instantaneous vibration. Furthermore, the segmented current increment strategy adopted by the welding process control module is divided into three stages: the first stage starts the arc at 40% of the rated current and lasts for 80 ms to establish the initial molten pool; the second stage linearly increases the current to 85% of the rated current within 120 ms to promote full melting of the base material; the third stage enters the constant current maintenance mode and performs dynamic fine-tuning within ±15% according to the melt depth feedback signal, with an adjustment cycle of 50 ms. Preferably, the penetration depth feedback signal comes from a dual-frequency laser confocal distance sensor, which is installed 15 mm behind the welding torch. The sampling frequency is 1 kHz, the measurement range is 0 to 5 mm, and the resolution is 1 μm. The measured value is processed by Kalman filtering and used as the closed-loop control input. Furthermore, the wire feeding speed adjustment adopts a feedforward-feedback composite control structure. The feedforward part presets the basic wire feeding rate based on the welding speed and current value by looking up a table. The feedback part corrects the deviation based on the visual inspection results of the weld pool width. The correction coefficient ranges from 0.85 to 1.15, and the adjustment response time is less than 80 ms. Furthermore, the pattern recognition model in the online quality assessment module is a lightweight convolutional neural network. Its input consists of a 256×256 pixel infrared thermal image sequence and a time-frequency diagram of acoustic emission signals. The output is the defect type classification result and confidence score. The model inference latency is less than 150 ms, and the classification accuracy has been verified to reach 98.7%. Preferably, when generating the rework path, the defect response processing module automatically selects the repair welding strategy based on the defect size and depth: when the defect depth is less than 1.5 mm, single-pass welding repair is used; when it is greater than or equal to 1.5 mm, multi-layer multi-pass welding is performed, with the thickness of each layer controlled at 1.2 mm ± 0.1 mm and the interlayer temperature maintained at 120℃ ± 10℃. On the other hand, a quick-joint welding control method for aluminum alloy pipe fittings includes the following steps: Step S110, acquiring spatial point cloud data of the end faces of two aluminum alloy pipe fittings to be welded in real time using a distributed three-dimensional scanning device, and performing coordinate alignment and noise filtering; Step S120, performing registration analysis on the preprocessed point cloud data using an improved iterative nearest point algorithm to calculate the 6-DOF docking deviation matrix between the two pipe fittings; Step S130, generating a compensating motion trajectory for fixture adjustment using a multi-objective optimization algorithm based on the docking deviation matrix and material specification parameters in the process database; Step S140, driving a 6-DOF parallel robot to perform spatial fine-tuning according to the compensating motion trajectory, monitoring the actual pose in real time and comparing it with the target trajectory to form closed-loop control; Step S150, when the docking deviation continues... When the welding progresses within the allowable tolerance zone for more than a set time threshold and the clamping state meets safety conditions, a welding start permit signal is issued; Step S160: The welding power supply is activated according to a segmented current increment strategy, and the wire feeding mechanism and welding torch moving device are started simultaneously to begin the formal welding operation; Step S170: During the welding process, the arc voltage, welding current, penetration depth, and molten pool image are collected in real time, and the wire feeding speed and welding speed are dynamically adjusted based on the feedback signal to maintain the stability of the molten pool; Step S180: After welding is completed, the thermal imaging sequence and acoustic emission signal of the weld area are collected immediately and input into the trained lightweight convolutional neural network for defect identification; Step S190: If the identification result shows that there is an excessive defect, a local repair welding path matching the defect features is automatically generated, and the repair welding unit is scheduled to perform the repair operation; Preferably, in step S110, the point cloud data is acquired by using an active light source to project Gray code and phase-shifted fringe patterns, combined with a camera array to capture deformed fringes, and high-density three-dimensional coordinates are calculated. The number of points in a single scan is not less than 500,000. Before point cloud registration, bilateral filtering and noise reduction processing are performed, and the kernel radius is twice the average point spacing. Furthermore, in step S130, the process database contains thermal conductivity, melting point, coefficient of linear expansion and recommended welding parameter combinations for different grades of aluminum alloys (such as 6061, 6082 and 7075). The optimization algorithm automatically matches the material type and calls the corresponding constraint conditions when generating the trajectory. In addition, in step S170, the molten pool image is acquired by a narrow-band filter vision sensor with a center wavelength of 650 nm, a bandwidth of 10 nm, and a frame rate of 200 fps. After the molten pool contour is extracted by the image segmentation algorithm, its area and aspect ratio are calculated as feedback basis for wire feeding adjustment. Preferably, the lightweight convolutional neural network structure in step S180 includes 5 convolutional layers and 2 fully connected layers, with convolutional kernel sizes of 7×7, 5×5, 3×3, 3×3, and 1×1, respectively. The activation function is a modified linear unit. During the training phase, the model uses a dataset containing 100,000 labeled samples for supervised learning, covering four typical defects: porosity, lack of fusion, edge bite, and cracks. Furthermore, in step S190, the repair welding unit is equipped with an independent wire feeding system and a pulsed MIG welding torch. The welding current control accuracy is ±1 A, the wire feeding speed adjustment resolution is 0.01 m / min, and the repair welding starting point is automatically offset from the original weld center by 1.5 mm to avoid repeated melting. In addition, a data buffer is set between the welding process control module and the online quality assessment module to store all process data within the most recent complete welding cycle, including timestamps, current and voltage values, pose signals, thermal images and acoustic signals. The data retention period is no less than 30 days, and it supports traceability query and offline analysis. Compared with the prior art, the present invention has the following beneficial effects: By using high-precision point cloud registration and closed-loop automatic adjustment mechanism, the assembly accuracy of aluminum alloy pipe fittings is improved to within ±0.1 mm, significantly reducing the incidence of welding defects caused by misalignment and uneven gaps. By adopting a segmented current increment strategy and real-time feedback control of penetration depth, the risk of burn-through in thin-walled pipes during the arc initiation stage is effectively suppressed, and the stability of the welding process is improved by more than 40%. Integrating online quality assessment and automatic rework functions enables real-time identification and closed-loop handling of welding defects, reducing manual inspection steps and shortening the overall production cycle by 25%. The system uses a deterministic communication protocol between its modules, and the control cycle is stable at the 2 ms level, which meets the real-time requirements of high-speed automated production lines. The method and system work together to support rapid switching between various aluminum alloy grades and pipe diameter specifications. It is highly versatile and suitable for high-end manufacturing fields such as aerospace and rail transportation. Attached Figure Description

[0006] Figure 1 This is a schematic diagram of the overall technical solution architecture of the quick-connect welding control method and system for aluminum alloy pipe fittings proposed in this invention; Figure 2This is a schematic diagram of the core principle framework of adaptive welding control based on point cloud registration and closed-loop feedback in this invention. Detailed Implementation

[0007] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.

[0008] Example 1, please refer to Figure 1 box Figure 2 This embodiment uses the butt welding task of a certain type of aluminum alloy conduit (material 6061-T6, outer diameter 80 mm, wall thickness 3.5 mm) in the aerospace field on an automated assembly line as an application scenario, specifically implementing a quick-connect welding control method and system for aluminum alloy pipe fittings. This scenario has extremely high requirements for the airtightness, mechanical strength, and deformation control of the welded joint. The allowable misalignment is no more than ±0.1 mm, the angular deviation is less than ±0.2°, and the entire process from fitting loading to welding completion must be completed within 120 seconds to ensure the production line's cycle time is matched. The system is deployed in a temperature-controlled cleanroom, with the ambient temperature maintained at 22±2℃ and the relative humidity controlled at 50%±10% to reduce the impact of thermal deformation and oxidation.

[0009] In step S110, the distributed 3D scanning device initiates spatial data acquisition. This device consists of four sets of multi-view structured light 3D scanning sensors, evenly distributed 90° circumferentially on a fixed support at the assembly station. Each sensor set includes a DLP structured light projection unit and two high dynamic range industrial CMOS cameras. The projection unit alternately projects Gray code sequences and three-step phase-shift sinusoidal fringe patterns at a frequency of 60 Hz. The camera array captures the deformation images of the fringe on the pipe end face at a synchronous frame rate of 200 Hz. A single complete projection-acquisition cycle generates no less than 520,000 3D coordinate points, covering an area extending axially from the pipe end face from 0 to 150 mm, meeting the complete perception requirements for the bevel geometry and deformation of the adjacent parent material. The acquired raw point cloud data is first subjected to pixel-level phase unpacking, using a four-step phase-shift algorithm combined with Gray code-assisted unwrapping to obtain a continuous phase field. Then, the phase values ​​are converted into spatial 3D coordinates through a camera-projector calibration parameter matrix. To eliminate outliers caused by surface reflections and stray ambient light, the system performs bilateral filtering for noise reduction. The filter kernel radius is set to twice the average point spacing (approximately 0.2 mm), and the color space standard deviation and coordinate space standard deviation are set to 0.15 and 0.1 mm, respectively, effectively suppressing noise while preserving geometric features. Point cloud data from all viewpoints are aligned using a pre-calibrated global coordinate system transformation matrix to form an ordered point cloud set in a unified coordinate system. The data refresh cycle is 5 ms to ensure dynamic tracking capabilities.

[0010] In step S120, the geometric matching analysis module receives the preprocessed point cloud data of the two pipe fitting end faces and executes the improved Iterative Closest Point (ICP) algorithm for registration analysis. The initial registration adopts a coarse alignment strategy based on normal vector distribution: first, the principal normal vector directions of the two pipe fitting end face point clouds are calculated, and preliminary attitude alignment is achieved by minimizing the deviation of the normal vector angle, controlling the initial rotation error within 5°. Subsequently, the fine registration stage begins, and the optimization objective function is defined as:

[0011] in, , These are the corresponding points in the source point cloud and the target point cloud, respectively. , Let be the unit normal vector of the corresponding point. , This is a local curvature estimate. and Let be the rotation matrix and translation vector to be found. and These are the coefficients of the normal vector consistency weight term and the curvature change constraint term, respectively, set to 0.6 and 0.2 in this embodiment. This objective function, based on the traditional point-to-point distance minimization, enhances the constraints on surface orientation continuity and local geometric feature consistency, effectively avoiding erroneous convergence in planar or low-curvature regions. In each iteration, the algorithm accelerates the nearest point search using a KD tree, updates the correspondence, and then uses singular value decomposition to solve for the optimal transformation matrix. The iteration termination condition is: the mean square error is less than 0.02 square millimeters, or the change in transformation parameters for three consecutive iterations is less than 0.005 mm (translation) and 0.005° (rotation). When either condition is met, the algorithm outputs the final 6-DOF docking deviation matrix, which is in the form of a homogeneous transformation matrix:

[0012] The translation component This represents the offset of the end face center along the X, Y, and Z axes, with the rotational component passing through Euler angles. This indicates the angular deviation around the three axes, with an accuracy of 0.01 mm and 0.05°, meeting the requirements for subsequent high-precision adjustments.

[0013] In step S130, the dynamic path planning module receives the docking deviation matrix and retrieves the physical parameters corresponding to the current pipe fitting material (6061-T6) from the process database: thermal conductivity 167 W / (m·K), melting point 582℃, and linear expansion coefficient 23.6×10⁻ 6 / ℃, and recommended welding parameter combinations (shielding gas: Ar+5%He, flow rate: 18 L / min, preheating temperature: 120℃). The path planning task is modeled as a multi-objective optimization problem, with optimization objectives including: minimizing the total energy consumption during fixture adjustment.

[0014] in and The first Torque and angular velocity of each joint; minimum adjustment time ; and maximum clamp load balance , The system provides real-time force control for each clamping point. Constraints include: continuous and differentiable path, acceleration not exceeding the square of 3 m / s², no interference between the clamp's motion envelope and surrounding equipment, and smooth attitude changes of the end effector. The optimization solver employs an improved non-dominated sorting genetic algorithm (NSGA-II) with a population size of 100, 50 generations, a crossover probability of 0.9, and a mutation probability of 0.1. The solution results in a nonlinear compensation trajectory, decomposed into five third-order Bézier curves, each defined by four control points to ensure continuity of position, velocity, and acceleration between segments. The interpolation period is 2 ms. The generated time-displacement-velocity-acceleration four-dimensional control sequence is encapsulated using the CANopen over EtherCAT protocol and transmitted to the automatic adjustment actuator via industrial real-time Ethernet.

[0015] In step S140, the automatic adjustment actuator receives the control sequence and begins spatial fine-tuning. This mechanism employs a 6-DOF Stewart parallel robot, whose six retractable legs are driven by servo electric cylinders. The end effector platform connects to the tubing via a flexible gripping unit. The flexible gripping unit consists of adaptive grippers driven by shape memory alloy, with a microporous suction cup structure covering the contact surface. It can precisely adjust the gripping force within the range of 200 N to 800 N, with a response time of less than 15 ms. During execution, high-precision encoders (resolution 0.001°) mounted at the leg joints and a six-dimensional force / torque sensor at the end effector (range ±1000 N, ±200 N·m) provide real-time feedback on the actual pose and force state. The dynamic path planning module compares the feedback data with the target trajectory and calculates the deviation vector. If the position deviation exceeds 0.02 mm or the attitude deviation exceeds 0.05°, the closed-loop correction mechanism is activated: a fine-tuning increment is inserted in the next 2 ms interpolation cycle. The adjustment amount is calculated according to the proportional-integral control law, with a proportional gain of 1.2 and an integral gain of 0.3, ensuring that the system converges the docking deviation to within ±0.05 mm and ±0.1° within 100 ms.

[0016] In step S150, the welding start determination module continuously monitors the docking status signal. The determination logic adopts a dual-condition triggering mechanism: First, the spatial docking deviation is less than ±0.1 mm in the X, Y, and Z translational directions and less than ±0.2° in the rotational deviation around the three axes; second, the clamp pressure fluctuation is less than 5% of the rated value (based on 500 N, the allowable fluctuation is ±25 N). The above conditions must be continuously met for 150 consecutive data sampling cycles (i.e., 300 ms) before a welding enable signal can be output. The monitoring data comes from the real-time feedback stream of the automatic adjustment actuator, with a sampling frequency of 200 Hz. The system is equipped with an anti-interference filtering window to avoid misjudgment due to mechanical vibration or instantaneous impact. Once the dual conditions are met, the welding start determination module sends an enable signal to the welding process control module through a dual channel of hard-wired and digital signals to start the welding preparation process.

[0017] In step S160, the welding process control module activates the welding power supply and executes a segmented current increment strategy. The first stage is the arc ignition stage, where the welding power supply outputs 40% of the rated current (set to 180 A), i.e., 72 A, for 80 ms. This stage uses a high-frequency pulse arc ignition method with a pulse frequency of 200 Hz and a duty cycle of 30%, ensuring a stable initial molten pool under low heat input and preventing burn-through of thin-walled pipes. The second stage is the melting and propagation stage, where the welding current is linearly increased from 72 A to 153 A (85% of the rated value) within 120 ms, with an increase slope of 0.675 A / ms. This stage is coordinated with the shielding gas being turned on 200 ms in advance (Ar + 5% He, 18 L / min) to form a stable gas shield. The third stage is the constant current maintenance stage, where the current stabilizes at 153 A, and a closed-loop feedback regulation mechanism is activated simultaneously. The wire feeding mechanism adopts a double active wire feeding roller structure. The wire material is ER5356 aluminum alloy welding wire with a diameter of 1.2mm. The basic wire feeding rate is preset to 5.2 m / min based on the welding speed (set to 0.45 m / min) and the current value.

[0018] In step S170, the system acquires multi-source feedback signals in real time to dynamically adjust the welding process. The penetration depth feedback signal is acquired by a dual-frequency laser confocal distance sensor, which is installed 15 mm behind the welding torch, with its optical axis at a 30° angle to the weld to avoid direct arc interference. The sensor uses alternating measurements of 1310 nm and 1550 nm wavelengths, with a sampling frequency of 1 kHz, a measurement range of 0 to 5 mm, and a resolution of 1 μm. The raw measurements are processed by a first-level median filter to remove impulse noise, followed by a second-level Kalman filter for state estimation. The state vector during the filtering process includes the penetration depth value and its first derivative. The process noise covariance is set to 1 × 10⁻⁸, and the observation noise covariance is set to 1 × 10⁻⁶. The output is a smooth penetration depth time series as the closed-loop control input. When the penetration depth deviation exceeds the set value of ±0.15 mm, the system dynamically fine-tunes the welding current within a range of ±15%, with an adjustment period of 50 ms. The adjustment amount is calculated according to the PD control law, with a proportional coefficient of 8 and a derivative coefficient of 0.5. Meanwhile, the molten pool image is acquired by a narrow-band filter vision sensor equipped with an optical filter with a center wavelength of 650 nm and a bandwidth of 10 nm, paired with a high-speed CMOS camera (200 fps, 1280×1024 resolution), capturing molten pool radiation through a side-view window. After image enhancement via adaptive histogram equalization, a modified U-Net network is used for semantic segmentation to extract the molten pool boundary and calculate its area and aspect ratio. When the molten pool area is less than the threshold of 8.5 square millimeters or the aspect ratio is greater than 2.3, it is determined that the molten pool has contracted or elongated, and the system initiates wire feed speed feedback adjustment. In the feedforward-feedback composite control structure, the feedforward part maintains a base rate of 5.2 m / min, and the feedback part calculates a correction coefficient based on the molten pool area deviation, with the range limited to between 0.85 and 1.15, and the adjustment response time is less than 80 ms. The welding speed is executed by the gantry welding torch moving device, whose servo motor encoder feeds back the position signal, forming a speed closed loop to ensure that the actual speed deviates from the set value by less than ±1%.

[0019] In step S180, the online quality assessment module is activated immediately after welding. The system acquires thermal imaging sequences and acoustic emission signals of the weld area. The infrared thermal imager is a 384×288 pixel focal plane array with a wavelength range of 7.5–13 μm and a frame rate of 50 Hz. It acquires thermal images of the cooling process within 60 seconds after welding. Each frame is resampled to 256×256 pixels after non-uniformity correction and bad pixel compensation. The acoustic emission sensor is a resonant piezoelectric probe with a center frequency of 150 kHz, installed on the pipe 50 mm from the weld, with a sampling frequency of 1 MHz and an acquisition time of 2 seconds. The signal is converted into a time-frequency map after bandpass filtering (100–300 kHz) (using short-time Fourier transform, window length of 1024 points, 50% overlap). The two types of data are then time-aligned and input into a lightweight convolutional neural network. The network structure consists of 5 convolutional layers and 2 fully connected layers: the first layer has a 7×7 kernel with a stride of 2 and outputs 64 channels; the second layer has a 5×5 kernel with a stride of 2 and outputs 128 channels; the third and fourth layers are 3×3 with a stride of 1, outputting 256 and 512 channels respectively; the fifth layer is 1×1 and outputs 512 channels. Global average pooling is then applied, followed by two fully connected layers (1024 and 512 neurons respectively), and finally, Softmax is used to output the classification probabilities of four types of defects (porosity, lack of fusion, undercut, and cracks). The activation function is Rectified Linear Unit (ReLU). The model underwent supervised learning on a training set containing 100,000 labeled samples, covering real defects under different pipe diameters, wall thicknesses, and welding parameters. Data augmentation included rotation, translation, and brightness perturbation. The inference latency was optimized to less than 150 ms, and the classification accuracy in field testing reached 98.7%. If the confidence score of any type of defect in the output exceeds 0.92, it is considered an out-of-range defect.

[0020] In step S190, if the online quality assessment module identifies a non-fusion defect with a depth of 1.8 mm, the defect response processing module immediately initiates the rework process. The system automatically selects a multi-layer, multi-pass welding strategy based on the defect depth (≥1.5 mm). First, a local machining path is generated based on the three-dimensional contour data of the defect area. The milling unit is controlled to remove the defect area down to the intact base material, with a machining depth of 2.0 mm and a width of 6.0 mm. Subsequently, the welding unit is scheduled to the repair position. This unit is equipped with an independent pulsed MIG welding torch, with welding current control accuracy of ±1 A and wire feed speed adjustment resolution of 0.01 m / min. In the welding path planning, the starting point is automatically offset by 1.5 mm from the center of the original weld bead to avoid repeated melting that could lead to grain coarsening. The welding employs a three-layer overlay, with each layer thickness controlled at 1.2 mm ± 0.1 mm. The interlayer temperature is monitored in real time via infrared thermography; if it falls below 110℃, local heating is initiated to 120℃ ± 10℃. The welding parameters for each layer are set independently: the first layer has a current of 140 A and a wire feed speed of 4.8 m / min; the second layer has 145 A and a wire feed speed of 5.0 m / min; the third layer has 150 A and a wire feed speed of 5.2 m / min, increasing layer by layer to compensate for heat accumulation. A quality assessment is triggered again after each weld repair until the defect is eliminated. All process data, including the original point cloud, deviation matrix, trajectory sequence, current-voltage curves, thermal images, acoustic signals, and defect determination results, are written to a data buffer and stored on a solid-state drive for at least 30 days. Traceability and offline statistical analysis are supported via the OPC UA interface.

[0021] Example 2 focuses on the welding of large-diameter aluminum alloy tubing (material 7075-T6, outer diameter 200 mm, wall thickness 8 mm) for rail transit vehicle underframes in a multi-product mixed-line production environment. It highlights the system's technical achievements in handling high-rigidity tubing, addressing differences in material thermal sensitivity, and enabling rapid process switching. Compared to Example 1, the core differences are: the geometric matching analysis module uses a fast registration algorithm based on key feature points instead of ICP to accommodate the surge in point cloud data caused by large workpieces; the dynamic path planning module introduces a material thermal deformation prediction model to compensate for initial deviations caused by welding preheating; and the welding process control strategy is adjusted to a four-stage current increment mode to meet the deep melting requirements of thick-walled tubing.

[0022] In step S110, the 3D scanning device still uses four sets of structured light sensors, but due to the increased size of the tube, the scanning range is expanded to the 0–300 mm axial region, the number of points scanned in a single scan is increased to 1.2 million, and the data refresh rate is reduced to 100 Hz to ensure real-time processing. The denoising process uses a combination of statistical outlier removal (SOR) and voxel mesh downsampling, with the voxel size set to 0.3 mm to control the point cloud density at a level that allows for efficient processing.

[0023] In step S120, the geometric matching analysis module no longer relies on full point cloud ICP, but instead adopts a registration process based on key points. First, ISS (Intrinsic Shape Signature) key point detection is performed on the point clouds of the two pipe fitting end faces, with parameters set to a neighborhood radius of 5 mm and a minimum strength difference of 0.02 m. Approximately 1200 key points are detected per end face. Then, the FPFH (Fast Point Feature Histograms) descriptor for each key point is calculated, with a dimension of 33. Descriptor matching is performed using a KD tree, and the RANSAC algorithm is used to filter inliers and estimate the initial transformation matrix. Finally, fine registration performs local ICP only within the neighborhood of the key points, significantly reducing the computational load. This algorithm reduces the registration time from 85 ms in Example 1 to 38 ms, meeting the fast cycle time requirements of mixed-line production.

[0024] In step S130, the dynamic path planning module, in addition to calling the 7075-T6 material parameters (thermal conductivity 130 W / (m·K), linear expansion coefficient 23.4 × 10⁻⁶ / ℃), also loads a thermal deformation prediction model. This model is built based on finite element offline simulation data, with the inputs being ambient temperature, pipe wall thickness, and preheating temperature (set to 180℃), and the output being the warping deformation field of the preheated rear end face. The system superimposes the predicted deformation field onto the current measured deviation matrix to generate an adjustment trajectory including pre-compensation, allowing the fixture to pre-adjust its attitude before preheating to counteract the effects of thermal deformation. In the multi-objective optimization, "minimizing the residual deviation after preheating" is added as a fourth objective to ensure that the docking accuracy remains within the tolerance zone after preheating.

[0025] In step S160, the welding process control module executes a four-stage current increment strategy: the first stage initiates the arc at 72 A (40%) for 80 ms; the second stage increases to 120 A (67%) within 100 ms to achieve stable penetration; the third stage increases to 171 A (95%) within 150 ms to promote deep penetration; and the fourth stage maintains 171 A and initiates feedback regulation. The penetration depth sensor measurement range is extended to 0–8 mm to accommodate thick walls, and the baseline wire feed rate is increased to 6.8 m / min. The remaining feedback control logic remains consistent with Example 1.

[0026] In step S180, the input size of the lightweight CNN model in the online quality assessment module is adjusted to 512×512 to accommodate large weld seam areas, and the network structure is adjusted to 6 convolutional layers, with an additional 3×3 convolutional layer added to improve feature extraction capabilities. The training dataset is supplemented with defect samples from thick-walled pipe fittings to ensure model generalization. This embodiment verifies the adaptability and scalability of the system under different material, size, and process requirements.

[0027] Example 3 focuses on the high-density array welding of small-diameter, thin-walled aluminum alloy tubing (material 6082, outer diameter 12 mm, wall thickness 1.0 mm) for cooling pipes in new energy vehicle battery packs. It highlights the system's technological achievements in preventing burn-through of ultra-thin-walled tubing, high-sensitivity detection of minute deviations, and collaborative optimization of multi-point continuous welding paths. Compared to the previous two examples, the substantial differences in this example are: the real-time data acquisition module uses line laser scanning instead of surface structured light to improve axial resolution; the welding process control module introduces molten pool oscillation frequency analysis as a stability criterion; and the dynamic path planning module executes a multi-tubing collaborative adjustment strategy.

[0028] In step S110, the real-time data acquisition module is equipped with two sets of line laser 3D scanners with a laser line width of 0.05 mm. These scanners scan along the pipe axis at 0.1 mm increments, acquiring 2000 cross-sectional contours in a single scan, which are then stitched together to form a high-resolution point cloud. The sampling resolution is increased to 0.02 mm, and the refresh rate reaches 500 Hz, ensuring the ability to capture minute vibrations in thin-walled pipes.

[0029] In step S170, in addition to the conventional melt depth and molten pool images, the system adds spectral analysis of the arc sound signal. Arc noise is collected via a microphone array, and the dominant frequency component is extracted using a Fast Fourier Transform. When the molten pool is stable, the dominant frequency is concentrated in the range of 800–1200 Hz; when a lack of fusion is observed, the frequency shifts down to 500–700 Hz. The system uses this frequency change as an auxiliary feedback signal, fusing it with the melt depth signal for judgment, thereby improving control robustness.

[0030] In step S130, the dynamic path planning module processes an array of 16 pipe fittings, requiring simultaneous optimization of the adjustment paths of all fixtures. The objective function includes a term to "minimize the relative displacement between adjacent pipe fittings" to avoid cascading disturbances caused by adjusting a single pipe. Path generation employs a distributed optimization algorithm, with each fixture node exchanging status information via real-time Ethernet to achieve coordinated movement. This embodiment verifies the system's control capabilities in complex multi-body assembly scenarios.

[0031] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A method for controlling a quick coupling weld of an aluminum alloy pipe fitting, characterized by, The method comprises the following steps: Real-time acquisition of spatial point cloud data of the end faces of two aluminum alloy pipes to be welded by a distributed three-dimensional scanning device, and coordinate alignment and noise filtering processing; Registration analysis of the pre-processed point cloud data is performed by using an improved iterative closest point algorithm, and a 6-DOF alignment deviation matrix between the two pipes is calculated; Based on the alignment deviation matrix and the material specification parameters in the process database, a multi-objective optimization algorithm is used to generate a compensation motion trajectory for clamp adjustment; A 6-DOF parallel robot is driven to perform spatial fine adjustment according to the compensation motion trajectory, and real-time monitoring of the actual pose is performed and compared with the target trajectory to form a closed-loop control; When the alignment deviation continuously and stably stays within the allowable tolerance band for more than a set time threshold, and the clamp pressure fluctuation amplitude is lower than the preset proportion of the rated value, a welding start permission signal is sent out; According to a segmented current increment strategy, a welding power source is activated, a wire feeding mechanism and a welding gun moving device are simultaneously started, and formal welding operation is started; During the welding process, arc voltage, welding current, penetration value and molten pool image are collected in real time, and the wire feeding speed and welding speed are dynamically adjusted based on the feedback signals to maintain the stability of the molten pool; Immediately after welding, a thermal imaging sequence and an acoustic emission signal of the weld area are collected, and input into a trained lightweight convolutional neural network for defect identification; if the identification result shows that there are excessive defects, a local repair welding path matching the defect characteristics is automatically generated, and a repair welding unit is dispatched to perform repair operation.

2. The aluminum alloy pipe fitting quick coupling welding control method according to claim 1, characterized by, The distributed three-dimensional scanning device adopts a multi-view structured light three-dimensional scanning sensor array, which is arranged on a circumferential fixed support of an assembly station, and the scanning range covers the axial extension area of the pipe end, the collected data is converted into spatial three-dimensional coordinates through phase unpacking and calibration parameter conversion, and is processed by bilateral filtering denoising to form an ordered point cloud set in a unified coordinate system.

3. The aluminum alloy pipe fitting quick coupling welding control method according to claim 1, characterized by, The improved iterative closest point algorithm introduces a normal vector consistency weight term and a curvature change constraint term in the optimization objective function, and realizes registration by minimizing the geometric distance, normal vector deviation and local curvature difference between the corresponding points, and the iteration termination condition is that the mean square error is lower than the preset threshold or the transformation parameter change amount of continuous multiple iterations is lower than the set accuracy.

4. The method of control of the welding of the aluminum alloy pipe fitting quick coupling according to claim 1, characterized by, The multi-objective optimization algorithm takes minimizing the adjustment energy consumption, the shortest adjustment time, the maximum clamp load balancing degree and the minimum residual deviation after preheating as the optimization objectives, generates a nonlinear compensation trajectory, which is decomposed into a smooth path composed of multiple Bezier curves, the interpolation period of each path is fixed, and a four-dimensional control sequence of time-displacement-velocity-acceleration is output.

5. The method of control of the welding of the aluminum alloy pipe fitting quick coupling according to claim 1, characterized in that, The 6-DOF parallel robot is connected to a flexible clamping unit at the end, the clamping force can be accurately adjusted within a set range, the actual pose and stress state are fed back in real time during the execution process through the joint encoder and the six-axis force / torque sensor, and are compared with the target trajectory, and when the deviation exceeds the preset threshold, a closed-loop correction mechanism is started, and a fine adjustment increment is inserted in the subsequent interpolation period to realize dynamic compensation.

6. The method of control of the welding of the aluminum alloy pipe fitting quick coupling according to claim 1, characterized in that, The trigger condition of the welding start permission signal comprises that the spatial docking deviation is within a preset tolerance range in the translation direction and the rotation direction, and the clamp pressure fluctuation amplitude is lower than 5% of the rated value, and the above conditions are continuously met in a plurality of sampling periods, and an anti-interference filtering window is arranged to avoid misjudgment caused by instantaneous vibration.

7. The method of control of the welding of the aluminum alloy pipe fitting quick coupling according to claim 1, characterized in that, The segmented current increment strategy comprises an arc striking stage, a melting expansion stage and a constant current maintaining stage, the arc striking stage adopts 40% of the rated current and cooperates with high-frequency pulse arc striking, the melting expansion stage linearly increases to 85% of the rated current within a set time, and the constant current maintaining stage dynamically adjusts the output current within a range of ±15% according to the molten depth feedback signal, the adjustment period is fixed, and the control law is based on a proportional-differential algorithm to calculate the adjustment amount.

8. The method of control of the welding of the aluminum alloy pipe fitting quick coupling according to claim 1, characterized in that, The molten depth value is obtained by a dual-frequency laser confocal ranging sensor, the sensor is installed at the side of the welding gun, the optical axis is at an angle with the weld, and the measurement signal is processed through median filtering and Kalman filtering to serve as the closed-loop control input; the molten pool image is collected by a narrowband filtering visual sensor, the image segmentation algorithm is used to extract the molten pool contour and calculate the area and the length-width ratio, and the molten pool image is used for feedback adjustment of the wire feeding speed.

9. The method of control of the welding of the aluminum alloy pipe fitting quick coupling according to claim 1, characterized in that, The lightweight convolutional neural network comprises a plurality of convolutional layers and fully connected layers, the input is an infrared thermal image sequence of 256*256 pixels and an acoustic emission signal time-frequency diagram, the output is a defect type classification result and a confidence score, the model completes supervised learning on a training set containing labeled samples, the inference delay is less than 150 ms, and the classification result is used to determine whether to start the repair process.

10. An aluminum alloy pipe fitting quick coupling weld control system, characterized by, The method comprises the steps of: a real-time data acquisition module is configured to acquire spatial pose data and contact edge information of the aluminum alloy pipe fittings during the assembly process; a geometry matching analysis module is connected with the real-time data acquisition module and is configured to perform alignment degree evaluation on the geometric features of the end faces of the two pipe fittings based on a point cloud registration algorithm to generate an initial docking deviation matrix; a dynamic path planning module receives the initial docking deviation matrix, combines a preset welding process parameter library, calculates a compensatory adjustment trajectory, and outputs an execution instruction; an automatic adjustment execution mechanism drives a clamp to real-time fine-tune the spatial position and attitude of the pipe fittings until the allowed docking tolerance range is reached in response to the output instruction of the dynamic path planning module; and a welding start determination module monitors the docking state signal, and triggers the welding equipment to enter a ready state when the docking deviation continuously and stably stays within a threshold value for more than a preset time window; a welding process control module starts the welding power supply according to the segmented current increment strategy after receiving the ready confirmation signal, and simultaneously monitors the arc stability and the molten depth feedback signal to dynamically adjust the wire feeding speed and the welding speed to maintain the dynamic balance of the molten pool; an online quality evaluation module collects a thermal imaging sequence and an acoustic emission signal of a weld area after the welding is completed, and determines whether there are incomplete fusion, porosity or crack defects through a pattern recognition model; a defect response processing module is connected with the online quality evaluation module, and generates a repair path and dispatches a repair welding unit to perform a local repair operation when an out-of-standard defect is identified; and the modules interact with each other through an industrial real-time Ethernet to ensure the deterministic transmission and synchronous execution of the control instructions.

Images