Vacuum laser in-situ cleaning and intelligent repair welding device and method

Abstract

The present application relates to the technical field of laser welding, and specifically discloses a vacuum laser in-situ cleaning and intelligent repair welding device and method, which comprises a vacuum welding cabin, a laser welding execution system, a micro-damage cleaning mechanism, a multi-modal visual perception system and an intelligent process control host; the micro-damage cleaning mechanism is used for sweeping the weld surface with constant force contact after welding is completed; the intelligent process control host automatically determines the corresponding repair mode according to the defect area type, controls the laser welding execution system to perform corresponding repair welding operation on each defect area according to the repair mode; after the repair welding is completed, the intelligent process control host rechecks the re-acquired weld surface image and feeds back the rechecking result and related parameters to the built-in welding process database. The present application can realize the whole-process closed loop from cleaning the surface to defect detection, intelligent decision of weld repair and final feedback of data.

Description

Technical Field

[0001] This invention relates to the field of laser welding technology, and more specifically to a vacuum laser in-situ cleaning and intelligent repair welding device and method. Background Technology

[0002] With the rapid development of the aerospace industry, extremely high requirements have been placed on the lightweighting and reliability of large structural components such as cabins. These components are typically made of thick aluminum alloy plates with a thickness ranging from 10mm to 60mm. Currently, vacuum laser welding, with its advantages of large penetration depth and high aspect ratio, is gradually becoming a key process for manufacturing complex structural components such as aluminum alloy cabins. However, during vacuum laser welding, due to the focused laser energy acting on the surface of the base material, the working temperature is extremely high. The intense evaporation of low-boiling-point elements in the aluminum alloy forms dense black ash on the weld surface, severely interfering with optical imaging and making it difficult to effectively monitor defects such as undercut and dents online. At the same time, traditional offline repair methods require repeated disruption of the vacuum environment and secondary clamping, which is not only inefficient but also makes it difficult to guarantee positioning accuracy. In addition, existing repair welding processes lack quantitative decision-making based on mathematical models, and relying on manual experience makes it difficult to achieve graded and precise repair for defects of different magnitudes. This easily induces secondary defects such as over-welding or under-filling, seriously restricting the manufacturing quality of thick aluminum alloy plates.

[0003] In recent years, intelligent welding repair technology has gradually become a research hotspot. However, the intelligent in-situ repair technology for aluminum alloy welding in a vacuum environment is still in the exploratory stage. It lacks a closed-loop execution link that automatically generates graded repair strategies based on detection results, and cannot realize a closed-loop automated process for defect repair from problem discovery to problem resolution.

[0004] Therefore, how to achieve a closed-loop process from surface cleaning to defect detection, intelligent decision-making for weld repair, and finally data feedback without breaking the vacuum environment is a technical problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0005] In view of the above problems, the present invention proposes a vacuum laser in-situ cleaning and intelligent repair welding device and method to overcome or at least partially solve the above problems.

[0006] To achieve the above objectives, the present invention adopts the following technical solution:

[0007] In a first aspect, the present invention provides a vacuum laser in-situ cleaning and intelligent repair welding device, comprising: a vacuum welding chamber, a laser welding execution system, a micro-destructive cleaning mechanism, a multimodal visual perception system, and an intelligent process control host; the laser welding execution system, the micro-destructive cleaning mechanism, and the multimodal visual perception system are all integrated inside the vacuum welding chamber and are communicatively connected to the intelligent process control host; The laser welding execution system is equipped with a laser for welding the workpiece to be welded; The micro-damage cleaning mechanism is used to sweep the weld surface with constant force after welding is completed. The multimodal vision perception system is used to scan the entire weld seam to obtain an image of the weld seam surface; The intelligent process control host is used to preprocess the weld surface image, generate several independent defect areas, and automatically determine the corresponding repair mode according to the type of defect area, and control the laser welding execution system to perform corresponding repair welding operations on each defect area according to the repair mode. The multimodal visual perception system is used to re-acquire images of the repaired area after the welding is completed; the intelligent process control host re-inspects the re-acquired weld surface image and calculates the flatness error E between the repaired weld surface and the weld surface reference plane; the defect feature value, the process parameters used, the cladding efficiency compensation coefficient under vacuum environment and the flatness error are used as a set of process parameter samples and fed back to the built-in welding process database.

[0008] Furthermore, the micro-damage cleaning mechanism includes a six-degree-of-freedom robotic arm, a cleaning head, and a vacuuming unit; The cleaning head uses a flexible rotating steel wire wheel, which is installed on the end flange of the six-degree-of-freedom robotic arm. It removes black and gray impurities from the weld surface through flexible contact. A telescopic module is set between the flange and the cleaning head. The intelligent process control host is used to drive the telescopic module to extend or retract along the normal direction of the weld surface, so that the insertion depth of the cleaning head is maintained within a preset range. The dust collection unit is coaxially wrapped around the outside of the cleaning head, forming an annular airflow channel to constrain and guide the black ash splashed during the cleaning process to the dust collection port; the dust collection port is located adjacent to the cleaning head and is connected to the outside of the vacuum welding chamber through a corrugated pipe.

[0009] Furthermore, the telescopic module incorporates a pressure sensor and a linear displacement compensation mechanism, the latter extending and retracting along the axial direction of the cleaning head; the pressure sensor collects the contact force between the cleaning head and the weld surface in real time. The intelligent process control host calculates the contact force in real time. With target constant force Force deviation between And based on force deviation Calculate the axial expansion and contraction compensation amount of the linear displacement compensation mechanism. Expansion / contraction compensation amount The calculation formula is: ;in, This is the proportional adjustment coefficient. This is the integral adjustment coefficient. This is the differential adjustment coefficient.

[0010] Furthermore, the intelligent process control host is also used to call the preset optimal indentation depth δ and target constant force based on the hardness value HV and oxide film thickness of the workpiece to be welded. The oxide film thickness is a process parameter pre-entered into the welding process database based on the aluminum alloy grade, surface treatment method, pre-welding inspection, and historical calibration data.

[0011] Furthermore, the multimodal visual perception system includes a visual detection unit and a structured light 3D measurement unit; the visual detection unit is used to acquire a two-dimensional texture image of the weld surface; the structured light 3D measurement unit is used to acquire a three-dimensional depth point cloud image of the weld surface.

[0012] Secondly, the present invention provides a vacuum laser in-situ cleaning and intelligent repair welding method, which is applicable to the device described above, and includes the following steps: S1. Vacuum the vacuum welding chamber. After vacuuming, control the laser welding execution system to perform initial welding on the workpiece to be welded. S2. After the initial welding is completed, the micro-damage cleaning mechanism is controlled to contact the weld surface with constant force to sweep the weld surface. S3. Use the multimodal vision perception system to scan the entire swept weld seam to obtain a surface image of the weld seam; S4. The intelligent process control host preprocesses the weld surface image to generate several independent defect areas; S5. The intelligent process control host automatically determines the corresponding repair mode based on the type of defect area, and controls the laser welding execution system to perform corresponding repair welding operations on each defect area according to the repair mode. S6. After the repair welding is completed, the multimodal vision perception system is controlled to re-acquire images of the repaired area; the intelligent process control host re-inspects the re-acquired weld surface image, calculates the flatness error E between the repaired weld surface and the weld surface reference plane, and takes the defect feature value, the process parameters used, the cladding efficiency compensation coefficient under vacuum environment and the flatness error as a set of process parameter samples, and feeds them back to the built-in welding process database.

[0013] Furthermore, in S4, the process of preprocessing the weld surface image by the intelligent process control host includes: By performing coordinate registration between the two-dimensional texture image and the three-dimensional depth point cloud image, a fused feature map of the weld area is obtained. Suspected defect pixels are extracted based on grayscale abrupt changes, depth of indentation, and surface curvature changes, and a defect mask is output through a pre-trained semantic segmentation model. Morphological closing operations and connected component labeling are performed on the defect mask to merge adjacent and connected defect pixels into several independent defect regions.

[0014] Furthermore, S5 includes: The critical missing volume value of the surface that cannot be filled by laser remelting was determined in advance through experiments. And the critical depth value at which the remelted molten pool can be effectively leveled. ; The intelligent process control host calculates the missing volume for each individual defect area. and maximum missing depth ; If the currently detected defect area meets the conditions and When the defect is identified as a minor defect, an in-situ remelting and repair welding process is generated, and the laser beam emitted by the laser welding execution system is controlled to remelt and smooth the defect surface in an "∞" shaped trajectory. If the currently detected defect area meets the conditions or If the defect is deemed significant, then a pre-built regression prediction model is used to generate laser power and welding speed for repair. The wire feeding speed is calculated according to the following formula: ;in, The cladding efficiency compensation coefficient under vacuum conditions and Used to compensate for metal evaporation losses. The cross-sectional area of ​​the welding wire. The length of the defect area along the weld direction; The laser welding execution system is controlled to perform in-situ remelting or filler wire welding operations sequentially on different defect areas according to the current repair mode.

[0015] Furthermore, the regression prediction model includes a laser power prediction model and a welding speed prediction model, and the construction and prediction processes respectively include: During the construction process, a training set is built using historical filler wire repair samples. Each training sample set includes the missing volume of the defect area. Maximum depth Defect area length Current vacuum level The base material grade code M, the actual laser power P used, and the welding speed for repair welding. The defect feature value X is used as the input feature vector, defined as follows: Laser power prediction models were established respectively. and welding speed prediction model ,in, and These are the model coefficients obtained using the weighted least squares method; During prediction, the missing volume of the current defect region is... Maximum depth Defect area length Current vacuum level The base material grade code M is input into the laser power prediction model and the welding speed prediction model, respectively, to obtain the laser power P and the repair welding speed. Define the current process parameters. .

[0016] Furthermore, S6 also includes: Preset standard flatness error threshold If the current group of process parameters meets The weights of this set of process parameter samples are updated according to the following formula to enhance the contribution of the current process parameter samples in the subsequent training of the regression prediction model; the update formula is: ,in, This indicates the weight of the sample of process parameters before the update. This indicates the updated weight of the set of process parameter samples. Update the learning rate for the weights; If the current group of process parameters meets the requirements Then, the current group of process parameter samples is added to the training sample set, and the coefficients of the regression prediction model are updated again using the weighted least squares method according to the weights of each group of process parameter samples. and At the same time, according to Update the cladding efficiency compensation coefficient under vacuum environment Where β represents the learning rate for updating the cladding efficiency compensation coefficient, Δh represents the average height deviation of the weld surface after repair welding relative to the weld surface reference plane, and k old k represents the cladding efficiency compensation coefficient before the update. new D represents the updated cladding efficiency compensation coefficient. max The maximum depth of the defect area is represented by the reference surface of the weld: the reference point set is obtained by selecting normal weld point clouds without depressions, undercuts or bulges from the outer periphery of the defect area and fitting them using the least squares method.

[0017] As can be seen from the above technical solution, compared with the prior art, the present invention has the following beneficial effects: 1. This invention introduces a micro-destructive cleaning mechanism that flexibly brushes the weld surface under constant force control. The indentation depth and target contact force are adjusted according to the base material hardness and preset oxide film thickness, ensuring the wire wheel maintains a stable indentation depth even when the weld surface height fluctuates. This effectively removes black ash / oxide obstructions generated during vacuum welding without damaging the base material. Simultaneously, it removes condensed black ash and oxides covering the weld surface, exposing the weld morphology and inhibiting secondary adhesion. Combined with a multimodal visual perception system, it reduces interference from high reflectivity and thermal radiation, achieving high-fidelity online acquisition of weld morphology, thus solving the problem of difficult online monitoring in vacuum environments.

[0018] 2. This invention preprocesses the weld surface image to generate multiple independent defect regions and automatically selects the appropriate repair mode for each defect region category, abandoning the traditional single fixed parameter welding repair mode. For minor defects, the stirring effect of the Lissajous trajectory is used for in-situ smoothing, avoiding weld beads caused by unnecessary filler wire; for larger defects, the filler wire amount is calculated based on the volume conservation model, achieving precise matching between the filler amount and the missing volume, significantly improving the consistency of the repair welding quality.

[0019] 3. This invention establishes a closed-loop process encompassing cleaning, inspection, decision-making, execution, and feedback. It calculates the flatness error using re-inspection data after welding repair and writes defect characteristic values, process parameters, vacuum compensation coefficients, and flatness errors into the welding process database. For samples that pass re-inspection, their sample weight is increased; for samples that fail re-inspection, the cladding efficiency compensation coefficient k is corrected based on the average height deviation. Furthermore, the model coefficients of the regression prediction model are updated using the weighted least squares method, enabling the model to have self-evolution capabilities and adapt to process fluctuations caused by changes in different batches of materials or the environment. Attached Figure Description

[0020] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0021] Figure 1 This is a schematic diagram of the vacuum laser in-situ cleaning and intelligent repair welding device provided in an embodiment of the present invention; Figure 2 This is a flowchart of the vacuum laser in-situ cleaning and intelligent repair welding method provided in this embodiment of the invention; Figure 3 This is a schematic diagram of the partitioned repair welding strategy process provided in an embodiment of the present invention; In the diagram, 1-welding vacuum chamber, 2-laser welding execution system, 3-wire feeder, 4-six-degree-of-freedom robotic arm, 5-cleaning head, 6-multimodal vision perception system, 7-intelligent process control host, 8-workpiece to be welded, and 9-base. Detailed Implementation

[0022] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0023] Example 1 like Figure 1 As shown in the figure, this embodiment discloses a vacuum laser in-situ cleaning and intelligent repair welding device, including: a vacuum welding chamber 1, a laser welding execution system 2, a micro-destruction cleaning mechanism, a multimodal visual perception system 6, and an intelligent process control host 7; the laser welding execution system 2, the micro-destruction cleaning mechanism, and the multimodal visual perception system 7 are all integrated inside the vacuum welding chamber 1 and are communicatively connected to the intelligent process control host 7.

[0024] The laser welding execution system 2 is equipped with a laser for welding the workpiece 8 placed on the base 9.

[0025] The micro-destructive cleaning mechanism is used to sweep the weld surface with constant force after welding is completed.

[0026] The multimodal vision perception system 6 is used to scan the entire weld seam to obtain an image of the weld seam surface.

[0027] The intelligent process control host 7 is used to preprocess the weld surface image, generate several independent defect areas, and automatically determine the corresponding repair mode according to the defect area type. It controls the laser welding execution system to perform the corresponding repair welding operation on each defect area according to the repair mode. The repair welding operation is either in-situ remelting or wire filler welding operation. When performing wire filler welding operation, the welding wire is fed through the wire feeder 3.

[0028] The multimodal vision perception system 6 is used to re-acquire images of the repaired area after the welding is completed; the intelligent process control host 7 re-inspects the re-acquired weld surface image and calculates the flatness error E between the repaired weld surface and the weld surface reference plane; the defect feature value, the process parameters used, the cladding efficiency compensation coefficient under vacuum environment and the flatness error are used as a set of process parameter samples and fed back to the built-in welding process database.

[0029] Specifically, the vacuum welding chamber can achieve a vacuum level of 10.-3 Pa; The laser welding execution system is installed on the top of the vacuum welding chamber and is equipped with a laser with a rated power of not less than 20kW, integrating off-axis wire feeding and oscillation functions.

[0030] The micro-damage cleaning mechanism includes a six-degree-of-freedom robotic arm 4, a cleaning head 5, and a dust collection unit.

[0031] The cleaning head 5 uses a flexible rotating steel wire wheel, which is installed on the end flange of the six-degree-of-freedom robotic arm 4. The bristles of the flexible rotating steel wire wheel are made of 304 stainless steel wire with a diameter of 0.1mm–0.3mm. It removes black and gray impurities from the weld surface through flexible contact without removing the base metal. A telescopic module is set between the flange and the cleaning head. The intelligent process control host is used to drive the telescopic module to extend or retract along the normal direction of the weld surface, so that the insertion depth of the cleaning head is maintained within the preset range.

[0032] The telescopic module incorporates a pressure sensor and a linear displacement compensation mechanism. The linear displacement compensation mechanism extends and retracts along the cleaning head's axial direction and includes a linear guide rail, ball screw, displacement encoder, and servo driver positioned along the cleaning head's axial direction. Before cleaning, a multimodal vision perception system acquires the local height of the weld and determines the weld surface normal. During cleaning, the pressure sensor collects the contact force between the cleaning head and the weld surface in real time. The intelligent process control host uses constant force to achieve the target. Based on the baseline, the contact force is calculated in real time. With target constant force Force deviation between And based on force deviation Calculate the axial expansion and contraction compensation amount of the linear displacement compensation mechanism. Expansion / contraction compensation amount The calculation formula is: ;in, This is the proportional adjustment coefficient. This is the integral adjustment coefficient. These are the differential adjustment coefficients. The three coefficients are obtained through contact force calibration tests and are used to adjust the immediate response of force deviation, eliminate cumulative deviation, and suppress the trend of change, respectively. Then, the linear module is driven to extend or retract along the normal direction of the weld surface, so that the wire wheel is pressed into the preset range, thereby compensating for the slight distance change between the flange surface and the weld surface and achieving constant force contact.

[0033] The dust collection unit is a ring-shaped negative pressure hood, coaxially wrapped around the outside of the cleaning head, forming a ring-shaped airflow channel to confine and guide the black ash splashed during cleaning to the suction port. The suction port is located adjacent to the cleaning head and connected to the outside of the vacuum welding chamber via a corrugated pipe, so as to promptly extract black ash and particles and reduce contamination and secondary adhesion within the chamber. The intelligent process control host calls the preset optimal indentation depth δ and target constant force based on the hardness value HV and oxide film thickness of the workpiece to be welded. The extension and retraction of the micro-damage cleaning mechanism are controlled; the oxide film thickness is a process parameter that is pre-entered into the welding process database based on the aluminum alloy grade, surface treatment method, pre-welding inspection and historical calibration data, so as to achieve adaptive cleaning parameter control for different aluminum alloy systems.

[0034] In this embodiment, for softer 5-series aluminum alloys, a shallower indentation depth δ=0.02mm and a smaller constant contact force F0=3N are automatically set to prevent excessive cutting by the wire wheel and scratches on the base material; for harder 2-series / 7-series aluminum alloys, the system automatically sets a deeper indentation depth δ=0.05mm and a larger constant contact force F0=8N to ensure that the dense oxide layer and stubborn black ash on the surface can be effectively destroyed and a stable and consistent cleaning effect can be obtained.

[0035] The multimodal vision perception system consists of two parts: a vision inspection unit and a structured light 3D measurement unit. The vision inspection unit transmits control signals via an EtherCAT bus to acquire two-dimensional texture images of the weld surface. The structured light 3D measurement unit uses a 405nm short-wavelength blue laser and transmits point cloud data via a GigE Vision bus to acquire three-dimensional depth point cloud images of the weld surface.

[0036] The intelligent process control host is located outside the vacuum welding chamber. It is an industrial control computer equipped with an Intel Core i9 processor and an NVIDIA RTX 4090 GPU. It has a built-in welding process database with self-evolving function and a big data analysis model (regression prediction model). It communicates with the laser welding execution system, the vacuum environment micro-destruction cleaning mechanism and the multimodal vision perception system through the EtherCAT industrial Ethernet bus and the GigE Vision vision bus to realize millisecond-level signal interaction between devices and generation of graded repair welding strategies.

[0037] Example 2 like Figure 2 As shown, this embodiment provides a vacuum laser in-situ cleaning and intelligent repair welding method, which is applicable to the above-mentioned device and includes the following steps: S1. The vacuum welding chamber is evacuated to a pressure ≤0.1Pa. In this embodiment, the vacuum pressure is 0.05Pa. The laser welding execution system is then controlled to perform initial welding on the workpiece to be welded. In this embodiment, the workpiece to be welded is a thick aluminum alloy workpiece with a thickness greater than or equal to 10mm. The laser power for initial welding is 8000W, and the welding speed is 1.5m / min.

[0038] S2. After the initial welding is completed, maintain a vacuum environment and control the micro-destructive cleaning mechanism to contact the weld surface with constant force. Use a rotating wire wheel to sweep the weld surface. The wire wheel is pressed into the surface to a depth of 0.02-0.05mm to remove the metal condensation black ash and oxides covering the surface.

[0039] Specifically, a six-degree-of-freedom robotic arm drives a flexible rotating steel wire wheel cleaning head to move to the weld area. The cleaning head bristles are made of 304 stainless steel with a diameter of 0.2mm. A coaxially mounted annular negative pressure shroud on the outside of the cleaning head opens simultaneously to form an annular airflow channel, constraining and guiding the black ash generated and splashed during brushing to the suction port. During the cleaning process, a pressure sensor in the telescopic module collects the contact force F(t) of the steel wire wheel in real time, and the control unit calculates the force based on the target constant force F0. And through a linear displacement compensation mechanism, the linear module is driven to perform minute expansion and contraction along the normal direction of the weld surface; when At that time, the cleaning head is pressed downwards to compensate. During this process, the cleaning head retracts along the normal direction to compensate, thereby maintaining constant force contact between the wire wheel and the weld surface and automatically compensating for minute gap changes. The robotic arm controls the wire wheel to contact the weld surface at a speed of 1800 rpm. The process control host automatically calls the cleaning parameters based on the hardness value of the base material and the preset oxide film thickness, setting the wire wheel pressing depth to δ=0.03mm and the constant force threshold within the corresponding range. This allows the wire wheel to gently brush away surface oxides and black ash without cutting the base metal, thereby removing black ash impurities and exposing the clean metal surface. The miniature negative pressure cyclone dust collection unit works simultaneously, with its suction port located adjacent to the cleaning head and connected to the external filter assembly of the chamber wall through a corrugated pipe, promptly sucking in the dust that is brushed off and entering the filter.

[0040] S3. The entire weld seam after sweeping is scanned using a multimodal vision perception system to obtain an image of the weld seam surface. The multimodal vision perception system includes a 405nm blue light laser profilometer and a high-resolution industrial camera. The 405nm blue light laser profilometer can suppress metal reflection interference. The intelligent process control host communicates with the actuator through the EtherCAT industrial Ethernet bus and is connected to the camera through the GigE Vision vision bus.

[0041] S4. The intelligent process control host preprocesses the weld surface image to generate several independent defect regions. After receiving the weld surface image, the intelligent process control host first performs grayscale normalization, filtering and noise reduction, and weld centerline extraction on the two-dimensional texture image. Then, it performs coordinate registration between the two-dimensional texture image and the depth point cloud image to obtain the fusion feature map of the weld region. Subsequently, it extracts suspected defect pixels based on grayscale abrupt changes, depth indentation, and surface curvature changes. It outputs a defect mask through a semantic segmentation model trained on historical weld samples. Finally, it performs morphological closing operations on the defect mask to remove holes and burrs. The connected component labeling algorithm is used to merge spatially adjacent and connected defect pixels into several independent defect regions.

[0042] S5, the intelligent process control host automatically determines the corresponding repair mode based on the type of defective area, and controls the laser welding execution system to perform corresponding repair welding operations on each defective area according to the repair mode. Specifically, this includes: 1) Pre-determine the critical missing volume value of the surface that laser remelting cannot fill through experiments. And the critical depth value at which the remelted molten pool can be effectively leveled. .

[0043] 2) The intelligent process control host calculates the missing volume for each independent defect area. and maximum missing depth .

[0044] In calculating the defect volume V d First, select the point cloud of normal welds without indentation, undercut, or protrusion around the defect area as the reference point set, and then use the least squares method to fit the reference surface Z0(x,y) of the weld surface. Straight welds can be fitted as follows: Welds with large curvature can be fitted as The reference plane is used to represent the ideal weld surface height after the defect is filled.

[0045] Then, each independent defect region S is discretized into N sampling micro-elements, and the defect depth is calculated for the i-th micro-element. Press again Calculate the defect volume, where ΔS i Let be the projected area of ​​the i-th sampled element; the maximum depth is calculated according to... The algorithm calculates and extracts the maximum depth of the defect region as an auxiliary feature vector. This algorithm can quantify the three-dimensional volume and depth of irregularly shaped defects, providing a data foundation for subsequent calculation of filler wire amount.

[0046] 3) Intelligent hierarchical decision-making logic: If the currently detected defect area meets the conditions and When a minor defect is identified, an in-situ remelting and repair welding process is initiated. The laser beam emitted by the laser welding execution system is controlled to remelt and smooth the defect surface in an ∞-shaped trajectory. Specifically, the laser beam follows the Lissajous curve equation for an ∞-shaped scan, the equation being... , Where x(t) and y(t) are the coordinates of the beam center; A is the oscillation amplitude; ω is the angular frequency; and φ is the phase difference. The high-frequency oscillation of the beam stirs the molten pool, enhancing the fluidity of the liquid metal, and surface tension is used to automatically smooth the weld seam, improving the weld surface formation.

[0047] If the currently detected defect area meets the conditions or If the defect is deemed significant, then a pre-built regression prediction model is used to generate laser power and welding speed for repair. The wire feeding speed is calculated according to the following formula: ;in, The cladding efficiency compensation coefficient under vacuum conditions and Used to compensate for metal evaporation losses. The cross-sectional area of ​​the welding wire. The length of the defect area along the weld direction.

[0048] The generation of filler wire repair welding process parameters adopts a regression prediction model based on incremental learning. The regression prediction model includes a laser power prediction model and a welding speed prediction model. When constructing the model, a training set is built using historical filler wire repair welding samples. Each training sample includes the missing volume of the defect area. Maximum depth Defect area length Current vacuum level The base material grade code M, the actual laser power P used, and the welding speed for repair welding. The defect feature value X is used as the input feature vector, defined as follows: In this context, the first '1' in the vector represents a constant term, used to introduce the intercept term in the regression prediction model, and does not correspond to specific physical characteristics of the defect. Laser power prediction models are then established separately. and welding speed prediction model ,in, and These are the model coefficients obtained using the weighted least squares method.

[0049] During prediction, the missing volume of the current defect region is... Maximum depth Defect area length Current vacuum level The base material grade code M is input into the laser power prediction model and the welding speed prediction model, respectively, to obtain the laser power P and the repair welding speed. Wire feeding speed v s Then, by applying the volume conservation formula mentioned earlier... The calculated parameters are then input as constraints into the laser welding execution system to define the current process parameters. .

[0050] 4) Generate a complete sequence of repair welding instructions, and drive the laser welding execution system via the EtherCAT bus to sequentially perform in-situ remelting or filler wire repair welding operations on different defect areas according to the current repair mode. For example... Figure 3 As shown, first move to region 1 to perform remelting, then move to region 2 to perform wire filling.

[0051] The following example illustrates S5: Suppose two defect regions are detected. Region 1 has a 10mm long area, and the missing volume V is calculated by integration. d =0.5mm³, maximum depth D max =0.2mm.

[0052] The detection results for region 2 are as follows: a region with a length of 15mm was found. After integration calculation, the missing volume V is... d =18.0mm³, maximum depth D max =1.5mm.

[0053] Preset volume threshold V th =2.0mm³, depth threshold D th =0.5mm.

[0054] For region 1: Determine and The defect was determined to be minor. An in-situ remelting process was implemented: laser power 3000W, no filler wire. The scanning trajectory utilized the Lissajous equation. , This creates an "∞" shaped light spot trajectory with an amplitude of 2mm and a frequency of 75Hz, using the Marangoni effect to smooth out the bite edges.

[0055] For region 2: Determine The defect was determined to be significant. A filler wire repair welding process was generated based on a regression prediction model: laser power 5000W, welding speed v. w =1000mm / min. Calculate the wire feed speed v. s Given that the welding wire diameter is 1.2 mm and the cross-sectional area is A w ≈1.13mm², defect length L d=15mm. Select vacuum compensation coefficient. Substitution That is, the wire feeding speed is set to approximately 1.2 m / min.

[0056] S6. After the welding is completed, control the multimodal vision perception system to re-acquire images of the repaired area; The intelligent process control host re-inspects the re-acquired weld surface image, calculates the flatness error E between the repaired weld surface and the weld surface reference plane, and takes the defect feature value X, the process parameter Y, the cladding efficiency compensation coefficient k under vacuum environment and the flatness error E as a set of process parameters, and feeds them back to the built-in welding process database.

[0057] Re-inspection of flatness error according to Calculate, where, To re-inspect the surface height after welding repair. The height of the weld surface reference plane is given by N, where N represents the number of valid sampling points in the repair area used for flatness error calculation. This represents the planar coordinates of the i-th valid sampling point in the coordinate system of the repair area. The average height deviation is further defined. Δh represents the average height deviation of the weld surface after repair welding relative to the reference surface of the weld surface; when Δh>0, it means that the surface after repair welding is too high relative to the reference surface, which is considered excessive filling; when Δh<0, it means that the surface after repair welding is too low relative to the reference surface, which is considered insufficient filling.

[0058] Preset standard flatness error threshold If the current group of process parameters meets The weights of this set of process parameter samples are updated according to the following formula: ,in, This indicates the weight of the sample of process parameters before the update. This indicates the updated weight of the set of process parameter samples. The learning rate is used to update the weights of the current group of process parameter samples, thereby enhancing the contribution of the current process parameters in the subsequent training of the regression prediction model. During the subsequent training of the regression prediction model, the weights of each group of process parameter samples are used as weighting factors in the weighted least squares method to solve for the model coefficients. Specifically, let X be the input feature vector of the j-th group of samples in the training set. j The actual process parameters output are y j The corresponding sample weight is w j The model is trained by minimizing the weighted sum of squared residuals. Obtain the model coefficients θ. Sample weights w jThe larger the value, the higher the proportion of the residual term corresponding to that sample group in the model training, and the greater its impact on the model coefficient update. Therefore, increasing the weight of the re-inspected qualified samples can make the subsequent regression prediction model prioritize fitting the process parameter samples with better repair effects.

[0059] If the current group of process parameter samples does not satisfy E≤E th If the filling is insufficient or excessive, it indicates that the repair area still has insufficient or excessive filling. The welding result is then judged based on the average height deviation Δh; when Δh < 0, it is considered insufficient filling, and when Δh > 0, it is considered excessive filling. The intelligent process control host adds the current group of process parameter samples to the training sample set and updates the coefficients θ of the regression prediction model using the weighted least squares method according to the weights of each group of process parameter samples. P and θ v ; At the same time, according to Update the cladding efficiency compensation coefficient under vacuum environment Where β represents the learning rate for updating the cladding efficiency compensation coefficient, Δh represents the average height deviation of the weld surface after repair welding relative to the weld surface reference plane, and Δh=Z ravg -Z 0avg Z ravg To re-inspect the average height of the weld surface after repair welding, Z 0avg k is the average height of the reference plane on the weld surface. old k represents the cladding efficiency compensation coefficient before the update. new D represents the updated cladding efficiency compensation coefficient. max The maximum depth of the defect area is represented by the reference surface of the weld: the reference point set is obtained by selecting normal weld point clouds without depressions, undercuts or bulges from the outer periphery of the defect area and fitting them using the least squares method.

[0060] Specifically, in this embodiment, the re-inspection result of region 2 shows a flatness error of +0.1mm, which is taken as the allowable error threshold. The defect was determined to be a slight convexity and met the requirements. The system will then use the defect characteristic value X and process parameters... The compensation coefficient k=1.15 and the re-inspection flatness error E are written into the database as process parameter samples, and then... Increase the weight of this sample; if subsequent samples are under-filled or over-filled, then... Update the cladding efficiency compensation coefficient k under vacuum conditions, add the corresponding process parameter samples to the training sample set, and update the coefficients θ of the regression prediction model again using the weighted least squares method. P and θ v ...

[0061] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other.

[0062] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A vacuum laser in-situ cleaning and intelligent repair welding device, characterized in that, include: The system comprises a vacuum welding chamber, a laser welding execution system, a micro-destructive cleaning mechanism, a multimodal visual perception system, and an intelligent process control host. The laser welding execution system, the micro-destructive cleaning mechanism, and the multimodal visual perception system are all integrated inside the vacuum welding chamber and are communicatively connected to the intelligent process control host. The laser welding execution system is equipped with a laser for welding the workpiece to be welded; The micro-damage cleaning mechanism is used to sweep the weld surface with constant force after welding is completed. The multimodal vision perception system is used to scan the entire weld seam to obtain an image of the weld seam surface; The intelligent process control host is used to preprocess the weld surface image, generate several independent defect areas, and automatically determine the corresponding repair mode according to the type of defect area, and control the laser welding execution system to perform corresponding repair welding operations on each defect area according to the repair mode. The multimodal visual perception system is used to re-acquire images of the repaired area after the welding is completed; the intelligent process control host re-inspects the re-acquired weld surface images and calculates the flatness error E between the repaired weld surface and the weld surface reference plane. The defect characteristic value, the process parameters used, the cladding efficiency compensation coefficient under vacuum environment and the flatness error are used as a set of process parameter samples and fed back to the built-in welding process database.

2. The vacuum laser in-situ cleaning and intelligent repair welding device as described in claim 1, characterized in that, The micro-damage cleaning mechanism includes a six-degree-of-freedom robotic arm, a cleaning head, and a vacuuming unit; The cleaning head uses a flexible rotating steel wire wheel, which is installed on the end flange of the six-degree-of-freedom robotic arm. It removes black and gray impurities from the weld surface through flexible contact. A telescopic module is set between the flange and the cleaning head. The intelligent process control host is used to drive the telescopic module to extend or retract along the normal direction of the weld surface, so that the insertion depth of the cleaning head is maintained within a preset range. The dust collection unit is coaxially wrapped around the outside of the cleaning head, forming an annular airflow channel to constrain and guide the black ash splashed during the cleaning process to the dust collection port; the dust collection port is located adjacent to the cleaning head and is connected to the outside of the vacuum welding chamber through a corrugated pipe.

3. The vacuum laser in-situ cleaning and intelligent repair welding device as described in claim 2, characterized in that, The telescopic module incorporates a pressure sensor and a linear displacement compensation mechanism, which extends and retracts along the axial direction of the cleaning head. The pressure sensor collects the contact force between the cleaning head and the weld surface in real time. The intelligent process control host calculates the contact force in real time. With target constant force Force deviation between And based on force deviation Calculate the axial expansion and contraction compensation amount of the linear displacement compensation mechanism. Expansion / contraction compensation amount The calculation formula is: ;in, This is the proportional adjustment coefficient. This is the integral adjustment coefficient. This is the differential adjustment coefficient.

4. The vacuum laser in-situ cleaning and intelligent repair welding device as described in claim 3, characterized in that, The intelligent process control host is also used to call the preset optimal indentation depth δ and target constant force based on the hardness value HV and oxide film thickness of the workpiece to be welded. The oxide film thickness is a process parameter pre-entered into the welding process database based on the aluminum alloy grade, surface treatment method, pre-welding inspection, and historical calibration data.

5. The vacuum laser in-situ cleaning and intelligent repair welding device as described in claim 1, characterized in that, The multimodal visual perception system includes a visual detection unit and a structured light 3D measurement unit; the visual detection unit is used to acquire a two-dimensional texture image of the weld surface; the structured light 3D measurement unit is used to acquire a three-dimensional depth point cloud image of the weld surface.

6. A vacuum laser in-situ cleaning and intelligent repair welding method, characterized in that, It is applicable to the apparatus as described in any one of claims 1-5, comprising the following steps: S1. Vacuum the vacuum welding chamber. After vacuuming, control the laser welding execution system to perform initial welding on the workpiece to be welded. S2. After the initial welding is completed, the micro-damage cleaning mechanism is controlled to contact the weld surface with constant force to sweep the weld surface. S3. Use the multimodal vision perception system to scan the entire swept weld seam to obtain a surface image of the weld seam; S4. The intelligent process control host preprocesses the weld surface image to generate several independent defect areas; S5. The intelligent process control host automatically determines the corresponding repair mode based on the type of defect area, and controls the laser welding execution system to perform corresponding repair welding operations on each defect area according to the repair mode. S6. After the repair welding is completed, the multimodal vision perception system is controlled to re-acquire images of the repaired area; the intelligent process control host re-inspects the re-acquired weld surface image, calculates the flatness error E between the repaired weld surface and the weld surface reference plane, and takes the defect feature value, the process parameters used, the cladding efficiency compensation coefficient under vacuum environment and the flatness error as a set of process parameter samples, and feeds them back to the built-in welding process database.

7. The vacuum laser in-situ cleaning and intelligent repair welding method as described in claim 6, characterized in that, In S4, the process of the intelligent process control host preprocessing the weld surface image includes: By performing coordinate registration between the two-dimensional texture image and the three-dimensional depth point cloud image, a fused feature map of the weld area is obtained. Suspected defect pixels are extracted based on grayscale abrupt changes, depth of indentation, and surface curvature changes, and a defect mask is output through a pre-trained semantic segmentation model. Morphological closing operations and connected component labeling are performed on the defect mask to merge adjacent and connected defect pixels into several independent defect regions.

8. The vacuum laser in-situ cleaning and intelligent repair welding method as described in claim 6, characterized in that, S5 include: The critical missing volume value of the surface that cannot be filled by laser remelting was determined in advance through experiments. And the critical depth value at which the remelted molten pool can be effectively leveled. ; The intelligent process control host calculates the missing volume for each individual defect area. and maximum missing depth ; If the currently detected defect area meets the conditions and When the defect is identified as a minor defect, an in-situ remelting and repair welding process is generated, and the laser beam emitted by the laser welding execution system is controlled to remelt and smooth the defect surface in an "∞" shaped trajectory. If the currently detected defect area meets the conditions or If the defect is deemed significant, then a pre-built regression prediction model is used to generate laser power and welding speed for repair. The wire feeding speed is calculated according to the following formula: ;in, The cladding efficiency compensation coefficient under vacuum conditions and Used to compensate for metal evaporation losses. The cross-sectional area of ​​the welding wire. The length of the defect area along the weld direction; The laser welding execution system is controlled to perform in-situ remelting or filler wire welding operations sequentially on different defect areas according to the current repair mode.

9. The vacuum laser in-situ cleaning and intelligent repair welding method as described in claim 8, characterized in that, The regression prediction model includes a laser power prediction model and a welding speed prediction model, and the construction and prediction processes respectively include: During the construction process, a training set is built using historical filler wire repair samples. Each training sample set includes the missing volume of the defect area. Maximum depth Defect area length Current vacuum level The base material grade code M, the actual laser power P used, and the welding speed for repair welding. The defect feature value X is used as the input feature vector, defined as follows: Laser power prediction models were established respectively. and welding speed prediction model ,in, and These are the model coefficients obtained using the weighted least squares method; During prediction, the missing volume of the current defect region is... Maximum depth Defect area length Current vacuum level The base material grade code M is input into the laser power prediction model and the welding speed prediction model, respectively, to obtain the laser power P and the repair welding speed. Define the current process parameters. .

10. The vacuum laser in-situ cleaning and intelligent repair welding method as described in claim 8, characterized in that, S6 also includes: Preset standard flatness error threshold If the current group of process parameters meets The weights of this set of process parameter samples are updated according to the following formula to enhance the contribution of the current process parameter samples in the subsequent training of the regression prediction model; the update formula is: ,in, This indicates the weight of the sample of process parameters before the update. This indicates the updated weight of the set of process parameter samples. Update the learning rate for the weights; If the current group of process parameters meets the requirements Then, the current group of process parameter samples is added to the training sample set, and the coefficients of the regression prediction model are updated again using the weighted least squares method according to the weights of each group of process parameter samples. and At the same time, according to Update the cladding efficiency compensation coefficient under vacuum environment Where β represents the learning rate for updating the cladding efficiency compensation coefficient, Δh represents the average height deviation of the weld surface after repair welding relative to the weld surface reference plane, and k old k represents the cladding efficiency compensation coefficient before the update. new D represents the updated cladding efficiency compensation coefficient. max The maximum depth of the defect area is represented by the reference surface of the weld: the reference point set is obtained by selecting normal weld point clouds without depressions, undercuts or bulges from the outer periphery of the defect area and fitting them using the least squares method.

Images