CN114453708A - Aluminum alloy welding defect online monitoring method - Google Patents
Aluminum alloy welding defect online monitoring method Download PDFInfo
- Publication number
- CN114453708A CN114453708A CN202210160241.7A CN202210160241A CN114453708A CN 114453708 A CN114453708 A CN 114453708A CN 202210160241 A CN202210160241 A CN 202210160241A CN 114453708 A CN114453708 A CN 114453708A
- Authority
- CN
- China
- Prior art keywords
- molten pool
- welding
- aluminum alloy
- image
- defects
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000003466 welding Methods 0.000 title claims abstract description 113
- 230000007547 defect Effects 0.000 title claims abstract description 56
- 238000000034 method Methods 0.000 title claims abstract description 53
- 229910000838 Al alloy Inorganic materials 0.000 title claims abstract description 44
- 238000012544 monitoring process Methods 0.000 title claims abstract description 26
- 239000001307 helium Substances 0.000 claims abstract description 23
- 229910052734 helium Inorganic materials 0.000 claims abstract description 23
- SWQJXJOGLNCZEY-UHFFFAOYSA-N helium atom Chemical compound [He] SWQJXJOGLNCZEY-UHFFFAOYSA-N 0.000 claims abstract description 23
- 230000011218 segmentation Effects 0.000 claims abstract description 23
- 230000003647 oxidation Effects 0.000 claims abstract description 16
- 238000007254 oxidation reaction Methods 0.000 claims abstract description 16
- 238000010801 machine learning Methods 0.000 claims abstract description 13
- 229910001338 liquidmetal Inorganic materials 0.000 claims description 16
- 238000013145 classification model Methods 0.000 claims description 12
- 230000003287 optical effect Effects 0.000 claims description 12
- 238000012545 processing Methods 0.000 claims description 11
- WFKWXMTUELFFGS-UHFFFAOYSA-N tungsten Chemical compound [W] WFKWXMTUELFFGS-UHFFFAOYSA-N 0.000 claims description 10
- 239000010937 tungsten Substances 0.000 claims description 10
- 229910052721 tungsten Inorganic materials 0.000 claims description 10
- 238000001914 filtration Methods 0.000 claims description 9
- 238000007781 pre-processing Methods 0.000 claims description 8
- 230000002159 abnormal effect Effects 0.000 claims description 6
- 230000004927 fusion Effects 0.000 claims description 5
- 239000011324 bead Substances 0.000 claims description 4
- 230000002146 bilateral effect Effects 0.000 claims description 4
- 238000006243 chemical reaction Methods 0.000 claims description 4
- 238000004140 cleaning Methods 0.000 claims description 4
- 238000005520 cutting process Methods 0.000 claims description 4
- 239000011159 matrix material Substances 0.000 claims description 4
- WYTGDNHDOZPMIW-RCBQFDQVSA-N alstonine Natural products C1=CC2=C3C=CC=CC3=NC2=C2N1C[C@H]1[C@H](C)OC=C(C(=O)OC)[C@H]1C2 WYTGDNHDOZPMIW-RCBQFDQVSA-N 0.000 claims description 3
- 238000002372 labelling Methods 0.000 claims description 3
- 230000000877 morphologic effect Effects 0.000 claims description 3
- 238000007637 random forest analysis Methods 0.000 claims description 3
- 238000013528 artificial neural network Methods 0.000 claims description 2
- 238000000605 extraction Methods 0.000 claims description 2
- 238000012706 support-vector machine Methods 0.000 claims description 2
- 238000010301 surface-oxidation reaction Methods 0.000 claims description 2
- 230000009466 transformation Effects 0.000 claims 1
- 238000001514 detection method Methods 0.000 abstract description 11
- 238000004519 manufacturing process Methods 0.000 abstract description 10
- 238000003860 storage Methods 0.000 abstract description 7
- 239000002699 waste material Substances 0.000 abstract description 3
- 238000013135 deep learning Methods 0.000 abstract description 2
- 239000007788 liquid Substances 0.000 abstract description 2
- 229910045601 alloy Inorganic materials 0.000 abstract 1
- 239000000956 alloy Substances 0.000 abstract 1
- 229910001095 light aluminium alloy Inorganic materials 0.000 abstract 1
- 238000010586 diagram Methods 0.000 description 7
- 230000003993 interaction Effects 0.000 description 6
- 238000011176 pooling Methods 0.000 description 6
- 230000035515 penetration Effects 0.000 description 5
- 230000000007 visual effect Effects 0.000 description 5
- XKRFYHLGVUSROY-UHFFFAOYSA-N Argon Chemical compound [Ar] XKRFYHLGVUSROY-UHFFFAOYSA-N 0.000 description 4
- 238000004422 calculation algorithm Methods 0.000 description 4
- 239000007789 gas Substances 0.000 description 4
- 238000012876 topography Methods 0.000 description 4
- 230000002457 bidirectional effect Effects 0.000 description 3
- 238000013527 convolutional neural network Methods 0.000 description 3
- 238000005516 engineering process Methods 0.000 description 3
- 238000011156 evaluation Methods 0.000 description 3
- 238000005457 optimization Methods 0.000 description 3
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 3
- YIXKIBQOLYDQFG-UHFFFAOYSA-N [He].[W] Chemical compound [He].[W] YIXKIBQOLYDQFG-UHFFFAOYSA-N 0.000 description 2
- 229910052786 argon Inorganic materials 0.000 description 2
- 230000015572 biosynthetic process Effects 0.000 description 2
- 239000000284 extract Substances 0.000 description 2
- 238000002844 melting Methods 0.000 description 2
- 230000008018 melting Effects 0.000 description 2
- 239000003380 propellant Substances 0.000 description 2
- 230000001681 protective effect Effects 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 241000287127 Passeridae Species 0.000 description 1
- 239000010953 base metal Substances 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 239000000498 cooling water Substances 0.000 description 1
- 238000002790 cross-validation Methods 0.000 description 1
- 238000003066 decision tree Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000010891 electric arc Methods 0.000 description 1
- 230000007613 environmental effect Effects 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 238000007667 floating Methods 0.000 description 1
- 238000003709 image segmentation Methods 0.000 description 1
- 239000011261 inert gas Substances 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 238000003908 quality control method Methods 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
- 238000012216 screening Methods 0.000 description 1
- 239000013589 supplement Substances 0.000 description 1
- 230000001360 synchronised effect Effects 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
- 238000012549 training Methods 0.000 description 1
- 238000012795 verification Methods 0.000 description 1
Images
Classifications
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B23—MACHINE TOOLS; METAL-WORKING NOT OTHERWISE PROVIDED FOR
- B23K—SOLDERING OR UNSOLDERING; WELDING; CLADDING OR PLATING BY SOLDERING OR WELDING; CUTTING BY APPLYING HEAT LOCALLY, e.g. FLAME CUTTING; WORKING BY LASER BEAM
- B23K9/00—Arc welding or cutting
- B23K9/095—Monitoring or automatic control of welding parameters
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B23—MACHINE TOOLS; METAL-WORKING NOT OTHERWISE PROVIDED FOR
- B23K—SOLDERING OR UNSOLDERING; WELDING; CLADDING OR PLATING BY SOLDERING OR WELDING; CUTTING BY APPLYING HEAT LOCALLY, e.g. FLAME CUTTING; WORKING BY LASER BEAM
- B23K9/00—Arc welding or cutting
- B23K9/095—Monitoring or automatic control of welding parameters
- B23K9/0953—Monitoring or automatic control of welding parameters using computing means
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B23—MACHINE TOOLS; METAL-WORKING NOT OTHERWISE PROVIDED FOR
- B23K—SOLDERING OR UNSOLDERING; WELDING; CLADDING OR PLATING BY SOLDERING OR WELDING; CUTTING BY APPLYING HEAT LOCALLY, e.g. FLAME CUTTING; WORKING BY LASER BEAM
- B23K9/00—Arc welding or cutting
- B23K9/16—Arc welding or cutting making use of shielding gas
- B23K9/167—Arc welding or cutting making use of shielding gas and of a non-consumable electrode
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B23—MACHINE TOOLS; METAL-WORKING NOT OTHERWISE PROVIDED FOR
- B23K—SOLDERING OR UNSOLDERING; WELDING; CLADDING OR PLATING BY SOLDERING OR WELDING; CUTTING BY APPLYING HEAT LOCALLY, e.g. FLAME CUTTING; WORKING BY LASER BEAM
- B23K9/00—Arc welding or cutting
- B23K9/32—Accessories
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B23—MACHINE TOOLS; METAL-WORKING NOT OTHERWISE PROVIDED FOR
- B23K—SOLDERING OR UNSOLDERING; WELDING; CLADDING OR PLATING BY SOLDERING OR WELDING; CUTTING BY APPLYING HEAT LOCALLY, e.g. FLAME CUTTING; WORKING BY LASER BEAM
- B23K2103/00—Materials to be soldered, welded or cut
- B23K2103/08—Non-ferrous metals or alloys
- B23K2103/10—Aluminium or alloys thereof
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Plasma & Fusion (AREA)
- Mechanical Engineering (AREA)
- Theoretical Computer Science (AREA)
- Arc Welding In General (AREA)
Abstract
An aluminum alloy welding defect online monitoring method belongs to the technical field of intelligent welding manufacturing. The invention aims at the requirement of on-line rapid monitoring of welding quality in the manufacturing of large aluminum alloy components of high-end equipment. The method comprises the steps of collecting a molten pool image by using a high-speed passive vision sensing system, simultaneously detecting a characteristic region which is not covered by an oxidation layer at the edge of the molten pool and the surface of the molten pool by using a semantic segmentation method based on deep learning, respectively extracting global characteristics and local texture characteristics of the molten pool morphology on the basis, and realizing early detection and identification of typical defects through a machine learning model. On one hand, the invention can realize early detection and early warning of welding defects and locate the defects, and avoid accidents such as shutdown and the like caused by waste products due to the defects; on one hand, the weld forming quality can be identified in real time, and a basis is provided for the on-line feedback control of the process parameters. The method can be widely applied to the helium arc welding process of thick plates and medium-thickness plate aluminum alloys, and is particularly suitable for the welding occasions of components such as large-scale aerospace aluminum alloy storage tanks, aluminum alloy liquid tanks and the like.
Description
Technical Field
The invention belongs to the technical field of intelligent welding manufacturing. Relates to an on-line monitoring method for aluminum alloy welding defects, which can be widely applied to the thick plate and medium plate aluminum alloy welding manufacture of large equipment such as a space propellant storage tank, an aluminum alloy liquid tank and the like.
Background
Tungsten-electrode helium arc welding is an important tungsten-electrode inert gas protection welding technology. Compared with the conventional argon arc welding, the helium arc welding has high temperature and concentrated heat under the same condition, so that the helium arc welding can obtain larger fusion depth and melting efficiency. Therefore, the tungsten electrode helium arc welding is widely applied to the welding of thick plates and aluminum alloys of medium plates in the manufacturing field of high-end equipment such as aerospace equipment, for example, the welding and manufacturing of large propellant storage tanks.
Because the aluminum alloy has the characteristics of easy oxidation, easy heat conduction, poorer helium arc stability than argon arc and the like, the weld forming quality consistency in actual welding production is difficult to guarantee, and weld forming defects such as undercut, snakelike weld bead, uneven weld and the like are easy to occur, and the problem is more prominent in thick plates, particularly non-flat welding positions. However, the existing rare research report on the defect mechanism of aluminum alloy slope-climbing TIG welding lacks theoretical support, the process is difficult to optimize from the root, and on the other hand, the actual production process has the interference of object states, equipment parameters, environmental conditions and other irreparable and unpredictable uncertain factors, so that the generation of welding defects cannot be completely avoided or inhibited. At present, the detection of tungsten electrode helium arc welding defects of medium and thick plate aluminum alloys still mainly depends on the traditional manual monitoring method of 'shooting in the welding process and visual observation' in the welding process and the nondestructive detection of welding seams with complicated and various steps and extremely time consumption after the welding is finished, the manufacturing efficiency of large aluminum alloy welding structures is seriously restricted, the mass production is limited, and efficient and reliable automatic and intelligent defect detection means are urgently needed.
The weld pool area is an important information source for quality control in the welding process, and by monitoring the dynamic evolution process of the weld pool in real time, on one hand, early detection and early warning of welding defects and defect positioning can be realized, and accidents such as waste generation and shutdown caused by the defects are avoided; on one hand, the weld forming quality can be identified in real time, and a basis is provided for the on-line feedback control of the process parameters. As a mainstream welding process monitoring means, the visual sensor can obtain abundant and visual dynamic information of the welding process. Early experiments and research results prove that a high-speed passive vision sensing system is used for collecting a molten pool image, a semantic segmentation method based on deep learning is adopted for simultaneously detecting a characteristic region which is not covered by an oxidation layer at the edge of the molten pool and the surface of the molten pool, the global characteristic and the local texture characteristic of the molten pool are respectively extracted on the basis, and early detection and identification of typical aluminum alloy tungsten-electrode helium arc welding defects can be realized through a machine learning model.
The prior art documents and patent retrieval find that Chinese invention patent with patent application number of 201910288591.X, namely a laser welding forming defect prediction classification method based on sparse representation, discloses a laser welding forming defect prediction classification method based on sparse representation, a molten pool image is collected, molten pool parameters are obtained through an industrial computer and are processed and analyzed, and the method can accurately realize the prediction classification of welding defects according to the relation between the geometric characteristic parameters of the molten pool and the forming quality of welding seams; chinese invention patent 202011556910.X discloses a penetration identification method based on the fusion of geometric and textural features of a bidirectional molten pool, which utilizes geometric features and textural features extracted from bidirectional synchronous molten pool images acquired by real-time sensing in the welding process to establish an RBF penetration identification model, utilizes the real-time molten pool images to pre-judge the penetration state on the basis, and feeds back the penetration state to a welding machine through a controller to realize penetration control; the invention discloses a weld defect classification method based on geometric features and an AdaBoost algorithm, which is disclosed in Chinese invention patent 'weld defect classification method based on geometric features and the AdaBoost algorithm' with the patent application number of 201910837875. X.
In summary, the existing welding defect monitoring technologies at home and abroad are not directed at aluminum alloy tungsten electrode helium arc welding defect detection, only relate to visual sensing for specific scenes and an automatic detection technology which utilizes a traditional image segmentation algorithm to extract conventional geometric characteristic parameters and combines a machine learning classification model, and no open report of an aluminum alloy welding defect online monitoring method based on semantic segmentation and molten pool combined characteristic parameters is seen at present.
Disclosure of Invention
The invention aims to supplement the defects of the prior art and provides an aluminum alloy welding defect on-line monitoring method to realize early detection and early warning of aluminum alloy tungsten-electrode helium arc welding defects and locate the defects.
In order to achieve the purpose, the invention adopts the following technical scheme:
an aluminum alloy welding defect on-line monitoring method is characterized by comprising the following steps:
1) aligning the axis of a camera light path of the high-speed passive vision sensing system to the aluminum alloy direct-current positive-connection tungsten electrode helium arc welding molten pool area, and collecting clear molten pool images in real time;
2) preprocessing the molten pool image;
3) performing semantic segmentation on the preprocessed molten pool image, and detecting a molten pool profile and a liquid metal area on the surface of the molten pool, which is not covered by an oxidation layer;
4) carrying out image processing on the liquid metal area on the surface of the molten pool, which is not covered by the oxidation layer, and extracting texture characteristics as local texture characteristic parameters of the molten pool; carrying out feature extraction on the molten pool profile, and extracting morphological features and average molten pool gray level features as global molten pool morphology feature parameters;
5) carrying out characteristic pretreatment fusion on the local texture characteristic parameters of the molten pool and the global characteristic parameters of the molten pool morphology to obtain a molten pool joint characteristic vector, inputting the molten pool joint characteristic vector into a machine learning classification model, identifying the forming quality of the tungsten-electrode helium arc welding seam by the machine learning classification model, and outputting a classification result;
6) and judging whether the classification result belongs to the abnormal welding seam for N times continuously through software, and performing defect labeling and alarm prompting on the related images.
In the technical scheme, the high-speed passive vision sensing system in the step 1) comprises an industrial camera and an optical filter; the dynamic range of the industrial camera is not lower than 60db, the molten pool area is shot from the upper part of the welding molten pool at an angle and a distance which enable the whole surface of the molten pool to be clearly imaged, and the image acquisition rate is not lower than 250 frames per second; the optical filter adopts a narrow-band optical filter with the central wavelength within the range of 610nm-670nm, is fixed at the front end of the lens of the industrial camera and is used for reducing arc interference in the welding process;
in the technical scheme, the molten pool area in the step 1) comprises a molten pool surface oxidation layer and a liquid metal area which is not covered by the oxidation layer on the molten pool surface;
in the above technical solution, the pretreatment in step 2) includes the steps of: cutting the molten pool image in the step 1) to reserve a molten pool area and remove edge noise; carrying out bilateral filtering processing on the molten pool image in the step 1) to remove image noise; stretching the contrast of the molten pool image in the step 1) by adopting a gamma conversion model to realize image enhancement, and ensuring that the texture of a liquid metal area on the surface of the molten pool, which is not covered by an oxidation layer, is clear and visible;
in the above technical solution, the semantic segmentation in step 3) is performed by using a semantic segmentation model deployed in advance in an image processing unit; the semantic segmentation model adopts a Deeplabv3+ network model improved based on a multi-level channel attention mechanism, and the multi-level channel attention mechanism module is fused with a Decoder decoding network of the semantic segmentation model and used for generating channel attention characteristics;
in the above technical solution, the texture features in step 4) include one or more of gray level co-occurrence matrix, wavelet transform, and Tamura texture;
in the above technical solution, the global characteristic parameters of the molten pool morphology in step 4) include a maximum width (W) of the molten pool, a maximum length (L) of the molten pool, a maximum area (S) of the molten pool, a rectangular degree (R) of a molten pool area, a circular degree (C), and an average gray level of the molten pool
In the above technical solution, the feature preprocessing in step 5) includes: data filling, cleaning, standardization and feature selection operation; the machine learning classification model is established by adopting one or more of a neural network, a support vector machine, a Bayesian network, a random forest, a limited Boltzmann machine and an extreme learning machine and adopting a data driving method;
in the above technical solution, the quality of the tungsten-electrode helium arc welding seam in step 5) is divided into four categories, including: undercut defects, serpentine weld beads, uneven welds, normal welds;
in the above technical solution, the value of N in step 6) is set by software program according to actual needs before step 1);
the invention has the following advantages and prominent technical effects: the method comprises the steps of collecting clear molten pool images by adopting a high-speed passive visual sensing system, detecting a molten pool contour and a liquid metal area on the surface of the molten pool, which is not covered by an oxidation layer, by utilizing semantic segmentation, extracting texture features closely related to the surface tension of the molten pool in the liquid metal area as local texture features, and carrying out real-time defect identification by utilizing a machine learning classification model in combination with the global morphology features of the molten pool contour. The invention can realize early detection and early warning of welding defects and positioning of the defects, avoid accidents such as waste generation, shutdown and the like caused by the defects, and provide a basis for identifying the welding seam forming quality in real time and performing on-line feedback control on process parameters. The method can be widely applied to tungsten electrode helium arc welding processes of thick plates and medium-thickness plate aluminum alloys, and is particularly suitable for key component welding occasions of space flight and aviation equipment, energy power equipment, logistics storage and transportation equipment and the like of large aluminum alloy storage tanks, storage tanks and the like.
Drawings
FIG. 1 is a flow chart of an online monitoring method for aluminum alloy welding defects, according to the invention;
FIG. 2 is a schematic structural diagram of a system for an aluminum alloy welding defect on-line monitoring method according to an embodiment of the invention;
in the figure: 1-a welding gun; 2-aluminum alloy weldment; 3-a bed head slewing mechanism; 4-a bed tail slewing mechanism; 5, a lathe bed; 6-camera fixing support; 7-industrial camera; 8, an optical filter; 9-high speed passive vision sensing system; 10-an image acquisition system; 11-an industrial personal computer; 12-human-computer interaction interface; 13-welding robot control cabinet; 14-a welding robot; 15-welding power supply; 16-protective gas; 17-a circulating water tank;
FIG. 3 is a schematic diagram of a network structure of a semantic segmentation model according to an embodiment of the present invention;
in the figure: 18-Encoder; 19-Decoder; 20, inputting a picture; 21-deep convolutional neural network; 22-high level semantic features; 23-1 × 1 convolution; 24-3 x3 convolution rate 6; 25-3 x3 convolution rate 12; 26-3 x3 convolution rate 18; 27-pooling images; 28-cavity pyramid pooling module; 29-five layers input characteristic diagram; 30-1 × 1 convolution; 31 — first feature map; 32-a multi-level channel attention mechanism module; 33, upsampling; 34-low level semantic features; 35-1 × 1 convolution; 36-low level semantic feature maps; 37-merging; 38-merging feature maps; 39-3 x3 convolution; 40, upsampling; 41-outputting the picture;
FIG. 4 is a schematic diagram of a result of semantic segmentation on a molten pool image and a molten pool topography global feature parameter according to an embodiment of the present invention;
in the figure, 42-the molten pool profile; 43-liquid metal areas of the bath surface not covered by an oxide layer; w is the maximum width of the molten pool; l-maximum length of molten pool; s, the maximum area of a molten pool; p-perimeter of the molten pool profile;
FIG. 5 is a schematic illustration of quality classification of tungsten-helium arc welding seam formation according to an embodiment of the present invention;
in the figure: 44-undercut defect; 45-serpentine welding pass; 46-uneven welds; 47-normal weld;
Detailed Description
The principles and operation of the present invention will be described in further detail below with reference to the accompanying drawings and examples.
FIG. 2 is a schematic structural diagram of a system for an aluminum alloy welding defect on-line monitoring method according to an embodiment of the invention; the system comprises a welding gun 1, an aluminum alloy weldment 2, a bed head swing mechanism 3, a bed tail swing mechanism 4, a bed body 5, a camera fixing support 6, an industrial camera 7, an optical filter 8, a high-speed passive vision sensing system 9, an image acquisition system 10, an industrial personal computer 11, a human-computer interaction interface 12, a welding robot control cabinet 13, a welding robot 14, a welding power supply 15, protective gas 16 and a circulating water tank 17; the welding gun 1 is a direct-current tungsten electrode helium arc welding gun; the electrode, helium and welding gun cooling water are stored in the welding gun 1; during welding, melting a base metal to form a molten pool by using electric arc heat generated between a tungsten electrode of the welding gun 1 and the aluminum alloy weldment 2; the welding gun 1 is rigidly connected to the welding robot 14; the aluminum alloy weldment 2 is fixed by the bed head swing mechanism 3 and the bed tail swing mechanism 4; the machine tail slewing mechanism 4 can move in the machine body 5; the rotation direction of the aluminum alloy weldment 2 rotates clockwise as shown in the figure, and the welding gun 1 does not move; the industrial camera 7 and the optical filter 8 form the high-speed passive vision sensing system 9; the high-speed passive vision sensing system 9 is fixedly connected to the welding gun body 1 through the camera fixing support 6; clear molten pool images shot by the high-speed passive vision sensing system 9 are transmitted to the industrial personal computer 11 through the image acquisition system 10; the industrial personal computer 11 and the human-computer interaction interface 12 are in bidirectional transmission; the human-computer interaction interface 12 is connected with the welding robot control cabinet 13 through a signal line; the welding robot control cabinet 13 controls the welding robot 14 to move through a signal line; the shielding gas 16 is helium; the welding robot control cabinet 13 is controlled by the welding power supply 15; the protective gas 16 and the circulating water tank 17 act on the welding gun 1 after being electrified by the welding power supply 15;
FIG. 1 is a flow chart of an online monitoring method for aluminum alloy welding defects, which comprises the following steps:
1) aligning the axis of a camera light path of the high-speed passive vision sensing system to the aluminum alloy direct-current positive-connection tungsten electrode helium arc welding molten pool area, and collecting clear molten pool images in real time; the high-speed passive vision sensing system 9 comprises the industrial camera 7 and the optical filter 8; in the embodiment, the industrial camera 7 shoots a molten pool area from the rear upper part of a welding molten pool, and the included angle between the axis of the camera light path and a welding gun is 45 degrees; the dynamic range of the industrial camera 7 is 60db, and the image acquisition rate is 250 frames per second; in the wavelength range of 610-690nm, the intensity of the arc is weak, and the interference of the arc can be effectively eliminated by adopting a band-pass filtering measure in the waveband; in this embodiment, the optical filter 8 is a narrow-band optical filter with a central wavelength in a range of 610nm to 670nm, and is fixed at the front end of the lens of the industrial camera 7;
2) clear molten pool images acquired by the high-speed passive vision sensing system 9 are transmitted to the industrial personal computer 11 for storage through the image acquisition system 10; preprocessing the molten pool image by using an image processing unit on the industrial personal computer 11, wherein the preprocessing comprises cutting, filtering and denoising and image enhancement operation; in the embodiment, the cutting is to cut the original molten pool image to 512x512 pixels, so as to reserve the molten pool area and remove edge noise; the filtering denoising adopts bilateral filtering, and the bilateral filtering can well keep the edge characteristics of the molten pool profile and simultaneously eliminate noise; the image enhancement adopts a gamma conversion model, the gray level of the image and the contrast of the stretched image can be expanded through the gamma conversion model, and the overall brightness of the molten pool image can be controlled by adjusting the model parameter y to ensure that the texture of a liquid metal area on the surface of the molten pool, which is not covered by an oxidation layer, is clear and visible;
3) performing semantic segmentation on the preprocessed molten pool image, and simultaneously detecting the molten pool outline 42 and a liquid metal area 43 on the surface of the molten pool, which is not covered by an oxidation layer; the semantic segmentation is performed by using a semantic segmentation model of an image processing unit which is deployed on the industrial personal computer 11 in advance; in this embodiment, the semantic segmentation model adopts a delabv 3+ network model improved based on a multi-level channel attention mechanism, and fig. 3 is a schematic network structure diagram of the semantic segmentation model according to the embodiment of the present invention; the semantic segmentation model consists of the Encoder 18 architecture and the Decode 19 architecture; an input picture 20 acquires high-level semantic features 22 and low-level semantic features 34 of an image through a deep convolutional neural network 21, the high-level semantic features 22 are transmitted to a cavity pyramid pooling module 28, the cavity convolutional layers and the pooling layers with different rates are adopted for convolution and pooling respectively to obtain a five-layer input feature map 29, the five-layer input feature map 29 is subjected to 1x1 convolution 30 to obtain a first feature map 31, and the first feature map 31 is input to a multi-level channel attention mechanism module 32 and is subjected to up-sampling to obtain a second feature map; the low-level semantic feature 34 is convolved with 1x1 to obtain the low-level semantic feature map 36; merging 37 said second feature map with said low level semantic feature map 36, resulting in a merged feature map 38; performing 3x3 convolution 39 and upsampling 40 on the merged feature map 38 to obtain an output picture 41; in this embodiment, the deep convolutional neural network 21 is a deep convolutional network with hole convolution; the hole pyramid pooling module 28 includes a fourth hole convolution and image pool 27; the rate rates of the hole convolution are respectively 1, 6, 12 and 18; the multilayer channel attention mechanism module 32 is fused in the Decoder19 framework and is used for acquiring channel attention mechanism characteristics, and the channel attention characteristics are fused with the five-layer input characteristic map 29 to obtain the second characteristic map; the upsampling 33, 40 is a bilinear difference upsampling;
4) performing image processing on the liquid metal area 43 on the surface of the molten pool, which is not covered by the oxidation layer, by using a real-time image processing unit in the industrial personal computer 11, and extracting texture features as local texture feature parameters of the molten pool; a real-time image processing unit in the industrial personal computer 11 extracts the characteristics of the molten pool profile 42, and extracts morphological characteristics and average molten pool gray level characteristics as global characteristics parameters of molten pool morphology; FIG. 4 is a schematic diagram of the result of semantic segmentation of the weld pool image and global characteristic parameters of the weld pool topography according to an embodiment of the present invention, including the weld pool profile 42 and the liquid metal region 43 of the weld pool surface not covered by the oxide layer; in this embodiment, the maximum width (W) of the molten pool, the maximum length (L) of the molten pool, the maximum area (S) of the molten pool, the rectangular degree (R) of the molten pool area, the circular degree (C), and the average gray level of the molten pool are extracted from the molten pool profile 42As a global characteristic parameter of the molten pool morphology; FIG. 4 shows a part of the global characteristic parameters of the molten pool topography and the values required by the calculation of the global characteristic parameters of the molten pool topography, which comprise: the maximum width of a molten pool, the maximum length of the molten pool, the maximum area of the molten pool and the contour perimeter of the molten pool; extracting texture characteristics from the liquid metal area 43 on the surface of the molten pool, which is not covered by the oxidation layer, as local texture characteristic parameters of the molten pool; in this embodiment, the texture feature adopts six-dimensional gray level co-occurrence matrix statistics, i.e. energy meanMean value of entropyMean contrast valueMean value of homogeneityMean of variance of textureCorrelation mean
5) A preprocessing module in the industrial personal computer 11 performs characteristic preprocessing fusion on the local texture characteristic parameters of the molten pool and the global characteristic parameters of the molten pool morphology to obtain a molten pool combined characteristic vector, inputs the molten pool combined characteristic vector into a machine learning classification model, identifies the forming quality of the tungsten-electrode helium arc welding seam by the machine learning classification model, and outputs a classification result; in this embodiment, the feature preprocessing includes data filling, cleaning, standardization, and feature selection operations; the data filling, cleaning and standardization are all carried out by calling a preproccussing library in sklern in python software; the normalization refers to normalizing the feature data in order to ensure that all features have the same contribution to identifying different forming qualities; in the embodiment, the extracted gray level co-occurrence matrix texture feature data and the global morphology feature data contain certain noise and data redundancy, and an optimal feature subset is selected for feature selection in order to improve the signal-to-noise ratio of the data; the feature selection adopts a filtering method (Fliter) to select an optimal feature subset; dividing the feature training set into four feature subsets based on four evaluation criteria of a filtering method (Fliter), and primarily screening redundant features in the feature subsets by using the evaluation criteria; further selecting a feature subset set screened based on the evaluation criteria by adopting sequential floating backward selection, inputting a verification set of the selected features into the constructed welding state prediction classification model, and testing the performance of the model; finally, selecting an optimal characteristic subset as a joint characteristic parameter of the molten pool based on the precision after 10 times of cross validation; in this embodiment, the machine learning model adopts a random forest optimization (SSA-RF) classification model optimized based on a sparrow optimization algorithm, which is trained in advance and issued by the industrial personal computer 11, and optimization parameters are a decision tree number and a minimum leaf point number; in this embodiment, the forming quality of the tungsten-electrode helium arc welding seam is divided into four categories, including: undercut defects, serpentine weld beads, uneven welds, normal welds; FIG. 5 is a schematic illustration of quality classification of tungsten-helium arc welding seam formation according to an embodiment of the present invention;
6) and judging whether the classification result belongs to the abnormal welding seam for N times continuously through the human-computer interaction interface 12, and carrying out defect labeling and alarm prompting on the related images. In this embodiment, the value N is set to 20 in software in advance according to the image acquisition frame rate of the industrial camera 7, and when the classification result shows that the abnormal weld joint belongs to 20 times continuously on the human-computer interaction interface 12, it is determined that the welding state at that time is abnormal; and marking corresponding defects of the image with the abnormal welding state according to the classification result and sending alarm information.
Claims (10)
1. An aluminum alloy welding defect on-line monitoring method is characterized by comprising the following steps:
1) aligning the axis of a camera light path of the high-speed passive vision sensing system to the aluminum alloy direct-current positive-connection tungsten electrode helium arc welding molten pool area, and collecting clear molten pool images in real time;
2) preprocessing the molten pool image;
3) performing semantic segmentation on the preprocessed molten pool image, and detecting a molten pool profile and a liquid metal area on the surface of the molten pool, which is not covered by an oxidation layer;
4) carrying out image processing on the liquid metal area on the surface of the molten pool, which is not covered by the oxidation layer, and extracting texture characteristics as local texture characteristic parameters of the molten pool; carrying out feature extraction on the molten pool profile, and extracting morphological features and average molten pool gray level features as global molten pool morphology feature parameters;
5) carrying out characteristic pretreatment fusion on the local texture characteristic parameters of the molten pool and the global characteristic parameters of the molten pool morphology to obtain a molten pool joint characteristic vector, inputting the molten pool joint characteristic vector into a machine learning classification model, identifying the forming quality of the tungsten-electrode helium arc welding seam by the machine learning classification model, and outputting a classification result;
6) and judging whether the classification result belongs to the abnormal welding seam for N times continuously through software, and performing defect labeling and alarm prompting on the related images.
2. The aluminum alloy welding defect online monitoring method according to claim 1, wherein the high-speed passive vision sensing system in the step 1) comprises an industrial camera and an optical filter; the dynamic range of the industrial camera is not lower than 60db, the molten pool area is shot from the upper part of the welding molten pool at an angle and a distance which enable the whole surface of the molten pool to be clearly imaged, and the image acquisition rate is not lower than 250 frames per second; the optical filter adopts a narrow-band optical filter with the central wavelength within the range of 610nm-670nm, is fixed at the front end of the industrial camera lens and is used for reducing arc interference in the welding process.
3. The method for on-line monitoring the welding defects of the aluminum alloy according to the claim 1 or 2, characterized in that the molten pool area in the step 1) comprises a molten pool surface oxidation layer and a liquid metal area of the molten pool surface which is not covered by the oxidation layer.
4. The method for on-line monitoring the welding defects of the aluminum alloy as recited in claim 1, wherein the pretreatment in the step 2) comprises the following steps: cutting the molten pool image in the step 1) to reserve a molten pool area and remove edge noise; carrying out bilateral filtering processing on the molten pool image in the step 1) to remove image noise; and (3) stretching the contrast of the molten pool image in the step 1) by adopting a gamma conversion model to realize image enhancement, and ensuring that the texture of a liquid metal area on the surface of the molten pool, which is not covered by an oxidation layer, is clear and visible.
5. The method for on-line monitoring the welding defect of the aluminum alloy according to claim 1, wherein the semantic segmentation in the step 3) is performed by utilizing a semantic segmentation model which is deployed in an image processing unit in advance; the semantic segmentation model adopts a Deeplabv3+ network model improved based on a multi-level channel attention mechanism, and the multi-level channel attention mechanism module is fused with a Decoder decoding network of the semantic segmentation model and used for generating channel attention characteristics.
6. The method for on-line monitoring the welding defect of the aluminum alloy according to claim 1, wherein the texture features of the step 4) comprise one or more of gray level co-occurrence matrix, wavelet transformation and Tamura texture.
7. The method for on-line monitoring the welding defects of the aluminum alloy according to the claim 1, wherein the global characteristic parameters of the molten pool morphology in the step 4) comprise a maximum width (W) of the molten pool, a maximum length (L) of the molten pool, a maximum area (S) of the molten pool, a rectangular degree (R) of a molten pool area, a circular degree (C) and an average gray level (g) of the molten pool.
8. The method for on-line monitoring the welding defects of the aluminum alloy as recited in claim 1, wherein the characteristic pretreatment in the step 5) comprises the following steps: data filling, cleaning, standardization and feature selection operation; the machine learning classification model is established by adopting one or more of a neural network, a support vector machine, a Bayesian network, a random forest, a limited Boltzmann machine and an extreme learning machine and adopting a data driving method.
9. The method for on-line monitoring of the welding defects of the aluminum alloy according to claim 1, wherein the forming quality of the tungsten electrode helium arc welding seam in the step 5) is divided into four types, including: undercut defects, serpentine weld beads, uneven welds, normal welds.
10. The method for on-line monitoring the welding defects of the aluminum alloy as recited in claim 1, wherein the value of N in the step 6) is set by a software program according to actual needs before the step 1).
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210160241.7A CN114453708A (en) | 2022-02-22 | 2022-02-22 | Aluminum alloy welding defect online monitoring method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210160241.7A CN114453708A (en) | 2022-02-22 | 2022-02-22 | Aluminum alloy welding defect online monitoring method |
Publications (1)
Publication Number | Publication Date |
---|---|
CN114453708A true CN114453708A (en) | 2022-05-10 |
Family
ID=81415387
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202210160241.7A Pending CN114453708A (en) | 2022-02-22 | 2022-02-22 | Aluminum alloy welding defect online monitoring method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN114453708A (en) |
Cited By (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN115018020A (en) * | 2022-08-08 | 2022-09-06 | 深圳市优米特新材料科技有限公司 | Improved deep learning titanium alloy processing equipment fault detection method |
CN115294132A (en) * | 2022-10-09 | 2022-11-04 | 江苏杰成电子科技有限公司 | Anodic oxidation layer defect assessment optimization method and system |
CN116597391A (en) * | 2023-07-18 | 2023-08-15 | 武汉纺织大学 | Synchronous on-line monitoring method for weld surface morphology and penetration state |
CN116900582A (en) * | 2023-07-19 | 2023-10-20 | 西咸新区大熊星座智能科技有限公司 | Welding robot with parameter prediction function |
CN117020361A (en) * | 2023-08-30 | 2023-11-10 | 鹤山市福维德智能设备有限公司 | Automatic detection device and detection method for mechanical manufacture |
CN117218188A (en) * | 2023-07-31 | 2023-12-12 | 深圳市大满包装有限公司 | Deep learning positioning optimization method for can body processing |
-
2022
- 2022-02-22 CN CN202210160241.7A patent/CN114453708A/en active Pending
Cited By (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN115018020A (en) * | 2022-08-08 | 2022-09-06 | 深圳市优米特新材料科技有限公司 | Improved deep learning titanium alloy processing equipment fault detection method |
CN115294132A (en) * | 2022-10-09 | 2022-11-04 | 江苏杰成电子科技有限公司 | Anodic oxidation layer defect assessment optimization method and system |
CN115294132B (en) * | 2022-10-09 | 2023-11-24 | 重庆景裕电子科技有限公司 | Anodic oxide layer defect evaluation optimization method and system |
CN116597391A (en) * | 2023-07-18 | 2023-08-15 | 武汉纺织大学 | Synchronous on-line monitoring method for weld surface morphology and penetration state |
CN116597391B (en) * | 2023-07-18 | 2023-09-19 | 武汉纺织大学 | Synchronous on-line monitoring method for weld surface morphology and penetration state |
CN116900582A (en) * | 2023-07-19 | 2023-10-20 | 西咸新区大熊星座智能科技有限公司 | Welding robot with parameter prediction function |
CN116900582B (en) * | 2023-07-19 | 2024-02-02 | 西咸新区大熊星座智能科技有限公司 | Welding robot with parameter prediction function |
CN117218188A (en) * | 2023-07-31 | 2023-12-12 | 深圳市大满包装有限公司 | Deep learning positioning optimization method for can body processing |
CN117020361A (en) * | 2023-08-30 | 2023-11-10 | 鹤山市福维德智能设备有限公司 | Automatic detection device and detection method for mechanical manufacture |
CN117020361B (en) * | 2023-08-30 | 2024-02-09 | 鹤山市福维德智能设备有限公司 | Automatic detection device and detection method for mechanical manufacture |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN114453708A (en) | Aluminum alloy welding defect online monitoring method | |
Xu et al. | Application of sensing technology in intelligent robotic arc welding: A review | |
CN109676135A (en) | A kind of laser gain material manufacture vision grey value difference on-line monitoring and bug repairing apparatus | |
Zhao et al. | Quality monitoring in wire-arc additive manufacturing based on cooperative awareness of spectrum and vision | |
CN113379740A (en) | VPPAW fusion in-situ real-time monitoring system based on perforation molten pool image and deep learning | |
CN116597391B (en) | Synchronous on-line monitoring method for weld surface morphology and penetration state | |
Cai et al. | Real-time identification of molten pool and keyhole using a deep learning-based semantic segmentation approach in penetration status monitoring | |
CN113554587B (en) | Deep learning-based molten pool image geometric feature extraction method and system | |
CN114723738A (en) | Precise welding microscopic monitoring method and system | |
CN113674206B (en) | Extraction method suitable for characteristic parameters of deep-melting K-TIG welding molten pool and keyhole entrance | |
Zhao et al. | Defect detection method for electric multiple units key components based on deep learning | |
CN116493735B (en) | Real-time tracking method for motion splash in Wanwave-level ultra-high power laser welding process | |
CN108067714A (en) | A kind of thin-walled circumferential weld termination quality on-line monitoring and defect positioning system and method | |
Liu et al. | Seam tracking system based on laser vision and CGAN for robotic multi-layer and multi-pass MAG welding | |
CN115908354A (en) | Photovoltaic panel defect detection method based on double-scale strategy and improved YOLOV5 network | |
Lin et al. | Intelligent seam tracking of an ultranarrow gap during K-TIG welding: a hybrid CNN and adaptive ROI operation algorithm | |
CN110414388A (en) | Hump and penetration on-line early warning method based on depth prediction network | |
CN109166088A (en) | Two waveband gray scale crater image fusion method based on non-down-sampled wavelet transformation | |
CN112529103A (en) | Fusion penetration identification method based on bidirectional molten pool geometric and textural feature fusion | |
He et al. | Weld seam profile extraction of T-joints based on orientation saliency for path planning and seam tracking | |
Chen et al. | Multi-exposure fusion for welding region based on multi-scale transform and hybrid weight | |
CN112801984B (en) | Weld joint positioning method based on countermeasure learning under laser vision system | |
CN117340459B (en) | Ultra-high power laser welding whole process monitoring method and system | |
Hong et al. | AF-FTTSnet: An end-to-end two-stream convolutional neural network for online quality monitoring of robotic welding | |
Fei et al. | Machine Vision Analysis of Welding Region and its Application to Seam Tracking in Arc Welding |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20220510 |
|
WD01 | Invention patent application deemed withdrawn after publication |