CN113751920A - Embedded device and method for detecting welding quality of lockhole TIG welding in real time - Google Patents
Embedded device and method for detecting welding quality of lockhole TIG welding in real time Download PDFInfo
- Publication number
- CN113751920A CN113751920A CN202110891305.6A CN202110891305A CN113751920A CN 113751920 A CN113751920 A CN 113751920A CN 202110891305 A CN202110891305 A CN 202110891305A CN 113751920 A CN113751920 A CN 113751920A
- Authority
- CN
- China
- Prior art keywords
- welding
- layer
- keyhole
- molten pool
- image
- 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.)
- Granted
Links
- 238000003466 welding Methods 0.000 title claims abstract description 230
- 238000000034 method Methods 0.000 title claims abstract description 51
- 230000035515 penetration Effects 0.000 claims abstract description 33
- 238000013528 artificial neural network Methods 0.000 claims abstract description 23
- 238000012544 monitoring process Methods 0.000 claims abstract description 18
- 238000005516 engineering process Methods 0.000 claims abstract description 12
- 230000009193 crawling Effects 0.000 claims abstract description 11
- 230000004927 fusion Effects 0.000 claims abstract description 7
- 238000013527 convolutional neural network Methods 0.000 claims abstract 5
- 238000012549 training Methods 0.000 claims description 27
- 230000008569 process Effects 0.000 claims description 25
- 230000006870 function Effects 0.000 claims description 15
- 239000000463 material Substances 0.000 claims description 12
- 238000011897 real-time detection Methods 0.000 claims description 10
- 238000010586 diagram Methods 0.000 claims description 9
- 230000000877 morphologic effect Effects 0.000 claims description 9
- 238000010606 normalization Methods 0.000 claims description 9
- 238000004891 communication Methods 0.000 claims description 6
- 150000001875 compounds Chemical class 0.000 claims description 6
- 239000004973 liquid crystal related substance Substances 0.000 claims description 6
- 238000012545 processing Methods 0.000 claims description 6
- 239000003086 colorant Substances 0.000 claims description 5
- 238000002844 melting Methods 0.000 claims description 5
- 238000012946 outsourcing Methods 0.000 claims description 5
- 238000005070 sampling Methods 0.000 claims description 5
- 239000011159 matrix material Substances 0.000 claims description 4
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 claims description 3
- 101100498818 Arabidopsis thaliana DDR4 gene Proteins 0.000 claims description 3
- 230000004913 activation Effects 0.000 claims description 3
- 239000000446 fuel Substances 0.000 claims description 3
- 238000011176 pooling Methods 0.000 claims description 3
- 230000003068 static effect Effects 0.000 claims description 3
- 230000001360 synchronised effect Effects 0.000 claims description 3
- 238000013519 translation Methods 0.000 claims description 3
- 238000012795 verification Methods 0.000 claims description 3
- 238000009432 framing Methods 0.000 claims description 2
- 238000007499 fusion processing Methods 0.000 claims description 2
- 238000003672 processing method Methods 0.000 claims description 2
- 230000002194 synthesizing effect Effects 0.000 claims description 2
- 238000004519 manufacturing process Methods 0.000 description 10
- 230000008859 change Effects 0.000 description 5
- 230000011218 segmentation Effects 0.000 description 4
- 230000007547 defect Effects 0.000 description 3
- 238000001514 detection method Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000005286 illumination Methods 0.000 description 2
- 230000006872 improvement Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000006467 substitution reaction Methods 0.000 description 2
- 230000015572 biosynthetic process Effects 0.000 description 1
- 230000007797 corrosion Effects 0.000 description 1
- 238000005260 corrosion Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 230000001788 irregular Effects 0.000 description 1
- 230000008018 melting Effects 0.000 description 1
- 239000002184 metal Substances 0.000 description 1
- 229910052751 metal Inorganic materials 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000001308 synthesis method Methods 0.000 description 1
- 238000012360 testing method 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
- B23K31/00—Processes relevant to this subclass, specially adapted for particular articles or purposes, but not covered by only one of the preceding main groups
- B23K31/12—Processes relevant to this subclass, specially adapted for particular articles or purposes, but not covered by only one of the preceding main groups relating to investigating the properties, e.g. the weldability, of materials
- B23K31/125—Weld quality monitoring
-
- 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
- B23K9/16—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
Landscapes
- Engineering & Computer Science (AREA)
- Mechanical Engineering (AREA)
- Physics & Mathematics (AREA)
- Plasma & Fusion (AREA)
- Quality & Reliability (AREA)
- Arc Welding In General (AREA)
Abstract
The invention discloses an embedded device and a method for detecting welding quality of lockhole TIG welding in real time. And determining a calibration exposure sequence through trial welding calibration, setting a multi-exposure mode of the CMOS camera, and monitoring a welding area through a multi-exposure fusion technology. And (3) utilizing an AI core board, segmenting the lock hole and the molten pool by using a CNN technology, extracting the characteristics of the size and the Hu invariant rectangular state, combining the characteristics of welding current and voltage as input, and identifying the penetration state in real time by using the constructed BP neural network. And uploading the image, the welding state and the early warning information to a remote server through a wireless module. The embedded device is suitable for lockhole TIG welding of the long welding seam of the curved surface of the medium plate, can be installed on a crawling robot, and realizes real-time identification and remote monitoring of welding quality.
Description
Technical Field
The invention relates to the technical field of welding monitoring and image processing, in particular to an embedded device and method for detecting the welding quality of lockhole TIG welding in real time.
Background
In recent years, with the development of advanced manufacturing technology and the continuous improvement of domestic market environment, the heavy machinery industry which plays an extremely important role in the manufacturing industry is rapidly developed, and the application demand of medium-thick metal plates (4.5 mm-25 mm) is increasing. At present, a medium plate welding structure is widely applied to industrial departments such as shipbuilding, large bridges, petrochemical industry, boiler containers, heavy machinery and the like. However, most of the large-sized weldments have large-curved-surface long welding seams, manual multi-layer and multi-pass welding can be adopted, the auxiliary processes are more, the quality stability of the welding seams is poor, the one-time welding yield is low, the rework amount is large, the production efficiency is low, the labor intensity of workers is high, and the production operation environment is severe. The current welding manufacturing process is difficult to deal with the long welding seam of the complex curved surface, and the improvement of the welding production and manufacturing efficiency of the medium and heavy plate in China is restricted.
The lockhole deep-melting TIG welding is a novel high-efficiency welding method for realizing large melting depth by utilizing a lockhole (keyhole) effect in the welding process, can realize one-pass penetration without groove opening, single-side welding and double-side forming, has good double-side weld formation, high welding quality, more stable lockhole TIG welding, has no splashing main defect of incomplete penetration or excessive penetration, and has great application prospect in the welding scene of medium plate materials.
However, the long welding line of the curved surface of the medium plate is in an irregular and random state due to the long stroke, the processing assembly and the processing deformation. The existing mature mode of an industrial control computer and a robot can only meet the manufacturing requirements of a common production line, but the mode is difficult to carry due to the thick volume of the robot, so that the real-time identification and remote monitoring of the welding quality in the moving process of a curved long welding seam are difficult to realize.
The invention patent CN 109719368B discloses a multi-information acquisition monitoring system and a method for a robot welding process, which utilize a main control computer with a wireless network card and a fixed industrial robot to realize real-time identification and remote monitoring of welding, and the method is suitable for welding application on a general production line. The method is difficult to be applied to the real-time identification and remote monitoring of the welding quality on the curved long welding seam which needs to move or crawl.
Disclosure of Invention
The invention aims to solve the defects in the prior art, and provides an embedded device and a method for detecting the welding quality of lockhole TIG welding in real time, which are suitable for real-time welding quality identification and remote monitoring of lockhole TIG welding on a long welding line of a medium plate curve.
The purpose of the invention can be achieved by adopting the following technical scheme:
the utility model provides an embedded device of lockhole TIG welding quality real-time detection, embedded device includes:
the CMOS camera 1 is connected with the AI core board 2 and used for acquiring a welding scene image of lockhole TIG welding in real time and transmitting image data to the AI core board 2;
the AI core board 2 is in data communication with other components of the embedded device and integrates a lockhole TIG welding quality real-time detection software package;
the LVDS liquid crystal touch screen 3 is connected with the AI core board 2 and is used for displaying and outputting local welding images in real time and realizing parameter setting input of an application interface including exposure time by using a touch function;
the mouse 4 is connected with the AI core board 2 and is used for assisting application software in clicking frames and inputting parameters;
the wireless module 5 is connected with the AI core board 2 and is used for transmitting image data, welding state and early warning information to a remote server;
the power supply module 6 is connected with the AI core board 2 and converts input voltage and/or current into various voltages to supply power to each component of the embedded device;
the welding current sensing module 7 is connected with the AI core board 2 and used for collecting and transmitting the welding current in the dynamic welding process of the robot in real time, and the Hall current sensor is used for communicating with the AI core board 2 through an RS232 serial port;
the welding voltage sensing module 8 is connected with the AI core board 2 and used for collecting and transmitting the welding voltage in the robot welding dynamic process in real time, and a Hall voltage sensor is used for communicating with the AI core board 2 through an RS232 serial port;
the crawling robot controller 9 is connected with the AI core board 2 and used for sending walking and welding instructions to the crawling robot and sending real-time pose information of a welding gun of the crawling robot to the AI core board 2;
the deep-melting lockhole TIG welding power supply 10 is connected with the AI core board 2 and is used for receiving current and voltage control signals sent by the AI core board 2;
the welding quality real-time detection method based on the embedded device comprises the following steps:
step S1: performing trial welding calibration, namely welding a same material according to current welding parameters, and respectively selecting clear exposure time of a welding gun, arc light, a lockhole and a molten pool by the CMOS camera 1 through acquiring images in a multi-exposure mode;
step S2: multi-exposure fusion, statically framing the areas of a welding gun, an arc light, a keyhole and a molten pool, adjusting the gains of 4 images to balance the brightness of the areas of the welding gun, the arc light, the keyhole and the molten pool, and synthesizing the 4 images into an image with clear details;
step S3: outputting a predicted feature map containing a keyhole and a molten pool by using the CNN technology with the image synthesized in the step S2 as an input;
step S4: according to the prediction characteristic diagram obtained in the step S3, according to RGB color values, respectively extracting keyhole pixel points and molten pool pixel points to obtain a keyhole image and a molten pool image, respectively performing binarization, wherein the foreground is 255 and the background is 0, firstly expanding and then corroding to obtain a maximum connected domain outline, calculating an outsourcing matrix, extracting the length and width of a keyhole and the molten pool according to the outsourcing matrix, and calculating morphological characteristics of Hu invariant moment according to the maximum connected domain outline;
step S5: and identifying the fusion penetration state according to the constructed BP neural network by taking the current welding voltage, the welding current, the length and the width of the lockhole and the molten pool and morphological characteristics as input.
Further, the size of the PCB of the AI core board 2 is 40mmx55mm, an RK1808K wide-temperature AI chip is adopted, the memory is 2GB DDR4 SDRAM, the memory is 8GB eMMC1 channels, the memory is provided with an LVDS liquid crystal screen interface, 1 channel gigabit ethernet interface, 1 channel USB3.0Host interface, 1 channel TF card interface, 1 channel audio/recording interface, 1 channel PCIE interface, 8 channels IO expansion interface, 4 channels UART interface, 3 channels SPI, 4 channels I2C, 4 channels analog acquisition ADC interface, 1 channel wireless WIFI module interface and 1 channel wireless 4G module interface;
the CMOS camera 1 has the functions of software triggering and exposure time setting, the exposure time range meets 100us-50ms, the dynamic range is larger than 60dB, and the frame rate is larger than 60 frames/s.
Further, the power module 6 includes a power management integrated circuit PMIC chip RK809, 1 configurable synchronous buck converter, 9 LDO regulators, two switches and a battery fuel gauge, and dynamically adjusts the output voltage of each DC-DC converter according to the operating state of the AI core board 2, so as to maximize the efficiency of the embedded device; the power supply module (6) converts the input voltage and current of 5V/3A into various voltages to supply power to all the components of the embedded device.
Further, the wireless module 5 adopts a communication module Huacheng ME909s 4G to transmit the thumbnail of the welding image, the welding speed, the welding current voltage, the length and width of the weld pool and the keyhole, the morphological characteristics and the penetration state to a remote server.
Further, the trial welding calibration in step S1 of the real-time detection method for welding quality is as follows: welding a same material according to current welding parameters, setting an exposure time sequence of a CMOS camera within the interval of 1100 us-50ms by taking 100us as a step, shooting each exposure image under the current welding parameters, manually selecting 4 exposure times corresponding to the exposure images of a welding gun, a welding arc, a keyhole and a molten pool respectively, setting the CMOS camera 1 to be in a multi-exposure mode, and sequentially setting the exposure times to be the 4 exposure times.
Further, the multi-exposure fusion process in step S2 of the real-time welding quality detection method is as follows: and selecting areas of a welding gun, an arc, a keyhole and a molten pool as an overlapping area for calculating gain in a static frame before welding, and optimizing the gain of 4 images according to the following formula:
in the formula (I), the compound is shown in the specification,
pixel brightness g corresponding to the overlapping parts of the welding gun, arc light, keyhole and molten pool0，g1，g2，g3Respectively representing the image gain, σI，σgLet σ be the standard deviation of the overlap region error and the standard deviation of the gain, respectivelyI＝20，σg＝0.1；
Fusing the multi-exposure images according to the following formula:
in the formula (I), the compound is shown in the specification,
is the brightness, Img ', of the pixel corresponding to the unadjusted image'iAnd the brightness of the pixel points corresponding to the unadjusted image is obtained.
Further, the input dimension of the CNN neural network constructed in step S3 of the welding quality real-time detection method is 512x512x3, and the output dimension is 512x512x 3;
the operation of the intermediate structure of the CNN neural network is a downsampling operation, a convolution operation and an upsampling operation;
wherein the down-sampling operation: the upper layer input is subjected to common convolution of a convolution kernel of 3x3, the number of layers is consistent with that of the upper layer input, and the stepping is 2, so that first output is obtained; the upper input is subjected to average pooling, and the step number is 2 to obtain a second output; stacking the first output and the second output; the output of the layer is obtained through batch normalization and linear rectification treatment, and the length and the width of the layer are halved through downsampling operation, but the layer number is doubled;
and (3) convolution operation: the upper layer input is subjected to expansion convolution of a convolution kernel of 3x3, the number of convolution layers is consistent with that of the upper layer input, the step is 1, the output of the layer is obtained through batch normalization and linear rectification, and the length, the width and the number of layers are not changed through convolution operation;
and (3) upsampling operation: the upper layer input is subjected to deconvolution of a convolution kernel of 3x3, the number of output layers is half of the number of the upper layer input layers, the step is 2, and the output of the layer is obtained through batch normalization and linear rectification processing; the up-sampling operation doubles the length and width dimensions, but halves the layer number;
the network structure comprises an input layer 512x512x3 downsampling operation to obtain a layer 1 output 256x256x6, a layer 1 output downsampling operation to obtain a layer 2 output 128x128x12, a layer 2 output downsampling operation to obtain a layer 3 output 64x64x24, a layer 3 convolution operation to obtain a layer 4 output 64x64x24, a layer 4 convolution operation to obtain a layer 5 output 64x64x24, a layer 5 upsampling operation to obtain a layer 6 output 128x128x12, a layer 6 upsampling operation to obtain a layer 7 output 256x256x6, and a layer 7 upsampling operation to obtain an output layer 512x512x 3.
Further, the training process of the CNN neural network is as follows:
the used training data are 5000 actual lockhole TIG welding images collected under the welding conditions of different welding speeds, welding currents, welding voltages and welding materials, the actual lockhole TIG welding images comprise states of incomplete penetration, proper penetration and excessive penetration, the labeled label images are three-channel mask images, and pixel points belonging to a background, a lockhole and a molten pool are endowed with different RGB colors respectively;
the training data were calculated as 7: and 3, dividing the training data into a training set and a verification set, simultaneously performing data enhancement on the training data in the training process, selecting data with a random probability of 25% in each training process to perform data enhancement operations including Gaussian noise, translation, scaling, mirror image, distortion and rotation, selecting the data batch size to be 8, the learning rate to be 0.001, selecting Adam by an optimizer, and selecting MSE by a loss function.
Further, in step S4 of the real-time welding quality detection method, the process of extracting the lengths and widths of the keyhole and the molten pool and the morphological characteristics according to the image processing method is as follows:
s41, respectively separating a keyhole image and a molten pool image according to RGB colors by the predicted characteristic diagram;
s42, respectively expanding and corroding the keyhole and the molten pool image, and then calculating a connected domain to obtain the maximum outline of the keyhole and the molten pool image;
s43, calculating the maximum of the keyhole and the molten pool respectivelyThe upper, lower, left and right boundaries of the large outline acquire the outer-wrapped rectangle, so that the length and width characteristics H of the lockhole are obtainedk、WkAnd the length and width characteristics H of the molten poolp、Wp；
S44, respectively calculating the invariant moment h of the keyhole Hu according to the maximum outlines of the keyhole and the molten poolk1，hk2，…hk7Constant torque h with molten pool hup1，hp2，…hp7。
Further, the BP neural network takes the size characteristics and the shape characteristics of the keyhole and the molten pool, and the current I and the voltage V measured by the welding current and voltage sensing module as input to form [ Hk，Wk，Hp，Wp，hk1，hk2，…hk7，hp1，hp2，…hp7，I，V]The BP neural network comprises two hidden layers, the number of the nodes of the first hidden layer is 128, the number of the nodes of the second hidden layer is 32, a Softmax function is selected as an activation function, and the number of iterations is set as 2000; the training data used by the BP neural network are 5000 actual keyhole TIG welding images acquired under different welding conditions including different welding speeds, welding currents, welding voltages and welding materials, and the actual keyhole TIG welding images comprise states of incomplete penetration, proper penetration and excessive penetration.
Compared with the prior art, the invention has the following advantages and effects:
1) the embedded device is suitable for real-time welding quality identification and remote monitoring of the lock hole TIG welding on the long welding line of the curved surface of the medium plate, can be installed on a crawling robot, and has the advantages of small volume, good flexibility and wide application range;
2) the invention adopts the common CMOS camera and multi-exposure fusion technology to replace the combination mode of an expensive high-dynamic camera and a filter, thereby being convenient, rapid and low in cost;
3) the CNN technology is used for replacing the traditional segmentation method, the method has the advantages of illumination resistance and false edge resistance, the network structure belongs to a lightweight class, and the network structure is applied to an RK1808K AI chip without an expensive industrial personal computer, so that a time-consuming segmentation algorithm has the characteristic of real-time performance;
4) the invention adopts the size characteristics and the form characteristics of the lock hole and the molten pool, and the current and the voltage measured by the welding current and voltage sensing module as the input characteristics for identifying the welding penetration state, and has higher accuracy in identification rate compared with the most using the size of the molten pool and considering the influence of the form characteristics.
Drawings
FIG. 1 is a structural diagram of an embedded device for detecting welding quality of lockhole TIG welding in real time in the embodiment of the invention;
FIG. 2 is a diagram of a method for detecting welding quality of keyhole TIG welding in real time in the embodiment of the invention;
FIG. 3 is a block diagram of a CNN neural network for dividing keyhole and molten pool in accordance with the embodiment of the present invention;
fig. 4 is a diagram of a BP neural network structure for identifying a penetration state disclosed in an embodiment of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are some, but not all, embodiments of the present invention. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
Examples
As shown in fig. 1, the embedded device for detecting the welding quality of keyhole TIG welding in real time comprises:
the CMOS camera is connected with the AI core board and used for acquiring a welding scene image of lockhole TIG welding in real time and transmitting image data to the AI core board; the CMOS camera has the functions of software triggering and exposure time setting, the exposure time range meets 100us-50ms, the dynamic range is larger than 60dB, and the frame rate is larger than 60 frames/s;
the AI core board is in data communication with other components of the embedded device and integrates a lockhole TIG welding quality real-time detection software package; the size of a PCB of the AI core board is 40mmx55mm, an RK1808K wide-temperature-level AI chip is adopted, an internal memory is 2GB DDR4 SDRAM, the memory is 8GBeMMC 1 paths, an LVDS liquid crystal screen interface is matched, 1 path of gigabit Ethernet interface, 1 path of USB3.0Host interface, 1 path of TF card interface, 1 path of audio/recording interface, 1 path of PCIE interface, 8 paths of IO expansion interface, 4 paths of UART interface, 3 paths of SPI, 4 paths of 12C, 4 paths of analog acquisition ADC interfaces, 1 path of WIFI module interface and 1 path of wireless 4G module interface;
the LVDS liquid crystal touch screen is connected with the AI core board and is used for displaying and outputting local welding images in real time and realizing parameter setting input of an application interface including exposure time by using a touch function;
the mouse is connected with the AI core board and is used for assisting application software in clicking frames and inputting parameters;
the wireless module is connected with the AI core board and is used for transmitting image data, welding state and early warning information to the remote server; the wireless module adopts a communication module Huawei ME909s 4G and transmits the welding image thumbnail, the welding speed, the welding current voltage, the length, the width, the morphological characteristics and the penetration state of a molten pool and a keyhole to a remote server.
The power supply module is connected with the AI core board and converts input voltage and/or current into various voltages to supply power to all components of the embedded device; the power module comprises a power management integrated circuit PMIC chip RK809, 1 configurable synchronous buck converter, 9 LDO regulators, two switches and a battery fuel meter, and the power module dynamically adjusts the output voltage of each DC-DC converter according to the working state of the AI core board so as to maximize the efficiency of the embedded device; the power supply module converts the input voltage and current of 5V/3A into various voltages to supply power for each component of the embedded device;
the welding current sensing module is connected with the AI core board and is used for collecting and transmitting the welding current in the dynamic welding process of the robot in real time, and the Hall current sensor is used for communicating with the AI core board through an RS232 serial port;
the welding voltage sensing module is connected with the AI core board and used for collecting and transmitting the welding voltage in the dynamic welding process of the robot in real time, and the Hall voltage sensor is used for communicating with the AI core board through an RS232 serial port;
the crawling robot controller is connected with the AI core board and used for sending walking and welding instructions to the crawling robot and sending real-time pose information of a welding gun of the crawling robot to the AI core board;
the TIG welding power supply for the deep-melting lockhole is connected with the AI core board and is used for receiving current and voltage control signals sent by the AI core board;
the process of the welding quality real-time detection method based on the embedded device is described as follows:
because the dynamic range of a common CMOS is generally not more than 80dB, and the welding process of K-TIG welding can generate extremely strong arc light due to the characteristic of large current, the CMOS is required to reach more than 140dB by observing a welding gun, the arc light, a lockhole and a molten pool at the same time, and the purpose can be achieved by an expensive high-dynamic camera and an optical filter. The method adopts a common CMOS camera multi-exposure mode + algorithm synthesis method to realize 140dB high dynamic, and can simultaneously observe a welding gun, arc light, a lockhole and a molten pool. The method comprises the following steps:
and (3) trial welding calibration, welding a same material according to current welding parameters, setting an exposure time sequence in a 0us-50ms interval of a CMOS camera by taking 100us as a step, shooting each exposure image under the current welding parameters, manually selecting 4 exposure times corresponding to a welding gun clear exposure image, an arc clear exposure image, a keyhole clear exposure image and a molten pool clear exposure image respectively, setting the CMOS camera to be in a multi-exposure mode, and sequentially setting the exposure times to be the 4 exposure times
And (3) fusion of multi-exposure images, wherein the areas of a welding gun, an arc, a keyhole and a molten pool are selected as an overlapping area for calculating gain in a static frame before welding, and the gains of 4 images are solved by optimizing the following formula:
in the formula (I), the compound is shown in the specification,
pixel brightness g corresponding to the overlapping parts of the welding gun, arc light, keyhole and molten pool0，g1，g2，g3Respectively representing the image gain, σI，σgLet σ be the standard deviation of the overlap region error and the standard deviation of the gain, respectivelyI＝20，σg＝0.1；
Fusing the multi-exposure images according to the following formula:
in the formula (I), the compound is shown in the specification,
is the brightness, Img ', of the pixel corresponding to the unadjusted image'iAnd the brightness of the pixel points corresponding to the unadjusted image is obtained.
The appearance of the penetration state on the image in the welding process is represented by characteristic changes of the lockhole and the molten pool, and the lockhole and the molten pool need to be accurately separated to extract the characteristics of the lockhole and the molten pool. And the actual welding image can change along with illumination and false edges, so that the traditional segmentation algorithm is difficult to stably extract. The current CNN technology is excellent in image feature extraction and can replace the traditional segmentation algorithm, but the CNN technology has the defects that the operation amount is extremely large by using a common processor and the CNN technology does not meet the real-time requirement. Therefore, a lightweight CNN network structure needs to be designed, and a special AI chip RK1808K is adopted to accelerate the running of the model so as to realize real-time performance.
The method comprises the steps of adopting a multi-exposure fused high-dynamic image as an input, outputting a prediction characteristic map containing a keyhole and a molten pool by using a CNN (computer-aided network) technology, wherein the input dimension is 512x512x3, and the output dimension is 512x512x 3;
the operation of the intermediate structure of the CNN neural network is a downsampling operation, a convolution operation and an upsampling operation;
wherein the down-sampling operation: the upper layer input is subjected to common convolution of a convolution kernel of 3x3, the number of layers is consistent with that of the upper layer input, and the stepping is 2, so that first output is obtained; the upper input is subjected to average pooling, and the step number is 2 to obtain a second output; stacking the first output and the second output; the output of the layer is obtained through batch normalization and linear rectification treatment, and the length and the width of the layer are halved through downsampling operation, but the layer number is doubled;
and (3) convolution operation: the upper layer input is subjected to expansion convolution of a convolution kernel of 3x3, the number of convolution layers is consistent with that of the upper layer input, the step is 1, the output of the layer is obtained through batch normalization and linear rectification, and the length, the width and the number of layers are not changed through convolution operation;
and (3) upsampling operation: the upper layer input is subjected to deconvolution of a convolution kernel of 3x3, the number of output layers is half of the number of the upper layer input layers, the step is 2, and the output of the layer is obtained through batch normalization and linear rectification processing; the up-sampling operation doubles the length and width dimensions, but halves the layer number;
the network structure comprises an input layer 512x512x3 downsampling operation to obtain a layer 1 output 256x256x6, a layer 1 output downsampling operation to obtain a layer 2 output 128x128x12, a layer 2 output downsampling operation to obtain a layer 3 output 64x64x24, a layer 3 convolution operation to obtain a layer 4 output 64x64x24, a layer 4 convolution operation to obtain a layer 5 output 64x64x24, a layer 5 upsampling operation to obtain a layer 6 output 128x128x12, a layer 6 upsampling operation to obtain a layer 7 output 256x256x6, and a layer 7 upsampling operation to obtain an output layer 512x512x 3.
The training process of the CNN neural network is as follows:
the used training data are 5000 actual lockhole TIG welding images collected under the welding conditions of different welding speeds, welding currents, welding voltages and welding materials, the actual lockhole TIG welding images comprise states of incomplete penetration, proper penetration and excessive penetration, the labeled label images are three-channel mask images, and pixel points belonging to a background, a lockhole and a molten pool are endowed with different RGB colors respectively;
the training data were calculated as 7: and 3, dividing the training data into a training set and a verification set, simultaneously performing data enhancement on the training data in the training process, selecting data with a random probability of 25% in each training process to perform data enhancement operations including Gaussian noise, translation, scaling, mirror image, distortion and rotation, selecting the data batch size to be 8, the learning rate to be 0.001, selecting Adam by an optimizer, and selecting MSE by a loss function.
The existing technical literature only predicts the penetration state through the size change of a lock hole and a molten pool, does not fully utilize the characteristics of a welding image, and is difficult to achieve higher recognition rate. In actual welding, the penetration state can also be reflected in the form change of the keyhole and the molten pool, the invention considers the size change of the keyhole and the molten pool and the form change of the keyhole and the molten pool, takes the current welding voltage, the welding current, the length and the width of the keyhole and the molten pool and the form characteristics as input, and identifies the penetration state according to the constructed BP neural network. The operation process is as follows:
respectively extracting pixel points of the lockhole and the molten pool from a predicted characteristic diagram comprising the lockhole and the molten pool output by the CNN technology according to RGB color values to obtain a lockhole image and a molten pool image, respectively carrying out binarization, wherein the foreground is 255 and the background is 0, carrying out expansion and then corrosion to obtain a maximum connected domain contour, respectively calculating the upper, lower, left and right boundaries of the maximum contour of the lockhole and the molten pool to obtain an outer covering rectangle, and further obtaining a lockhole length and width characteristic Hk、WkAnd the length and width characteristics H of the molten poolp、Wp(ii) a Respectively calculating the invariant moment h of the lockhole Hu according to the maximum outlines of the lockhole and the molten poolk1，hk2，…hk7Invariant h with the bath Hup1，hp2，…hp7。
The constructed BP neural network takes the size characteristics and the form characteristics of a lock hole and a molten pool and the current I and the voltage V measured by a welding current and voltage sensing module as input to form [ Hk，Wk，Hp，Wp，hk1，hk2，…hk7，hp1，hp2，…hp7，I，V]The BP neural network comprises two hidden layers, the number of the nodes of the first hidden layer is 128, the number of the nodes of the second hidden layer is 32, a Softmax function is selected as an activation function, and the number of iterations is set as 2000; it is composed ofThe training data used by the BP neural network are 5000 actual keyhole TIG welding images collected under different welding conditions including different welding speeds, welding currents, welding voltages and welding materials, and the actual keyhole TIG welding images comprise states of incomplete penetration, proper penetration and excessive penetration.
Compared with the method that the input feature prediction accuracy is 93.4% by the same data set test and only the size of the molten pool and the size of the lockhole are used as the input features, the prediction accuracy is 89.6%, and the feature identification accuracy is higher.
The above embodiments are preferred embodiments of the present invention, but the present invention is not limited to the above embodiments, and any other changes, modifications, substitutions, combinations, and simplifications which do not depart from the spirit and principle of the present invention should be construed as equivalents thereof, and all such changes, modifications, substitutions, combinations, and simplifications are intended to be included in the scope of the present invention.
Claims (10)
1. The utility model provides a lockhole TIG welds embedded device of welding quality real-time detection which characterized in that, embedded device includes:
the CMOS camera (1) is connected with the AI core board (2) and used for acquiring welding scene images of lockhole TIG welding in real time and transmitting image data to the AI core board (2);
the AI core board (2) is in data communication with other components of the embedded device and integrates a lockhole TIG welding quality real-time detection software package;
the LVDS liquid crystal touch screen (3) is connected with the AI core board (2) and is used for displaying and outputting local welding images in real time and realizing parameter setting input of an application interface including exposure time by using a touch function;
the mouse (4) is connected with the AI core board (2) and is used for assisting application software in clicking frame selection and inputting parameters;
the wireless module (5) is connected with the AI core board (2) and is used for transmitting image data, welding state and early warning information to the remote server;
the power supply module (6) is connected with the AI core board (2) and converts input voltage and/or current into various voltages to supply power to all the components of the embedded device;
the welding current sensing module (7) is connected with the AI core board (2) and is used for collecting and transmitting the welding current in the robot welding dynamic process in real time, and the Hall current sensor is used for communicating with the AI core board (2) through an RS232 serial port;
the welding voltage sensing module (8) is connected with the AI core board (2) and used for collecting and transmitting the welding voltage in the robot welding dynamic process in real time, and a Hall voltage sensor is used for communicating with the AI core board (2) through an RS232 serial port;
the crawling robot controller (9) is connected with the AI core board (2) and used for sending walking and welding instructions to the crawling robot and sending real-time pose information of a welding gun of the crawling robot to the AI core board (2);
the deep-melting lockhole TIG welding power supply (10) is connected with the AI core board (2) and is used for receiving current and voltage control signals sent by the AI core board (2);
the welding quality real-time detection method based on the embedded device comprises the following steps:
step S1: the method comprises the following steps of (1) trial welding calibration, wherein a same material is welded according to current welding parameters, and clear exposure time of a welding gun, arc light, a lockhole and a molten pool is respectively selected by a CMOS camera (1) through multi-exposure mode image acquisition;
step S2: multi-exposure fusion, statically framing the areas of a welding gun, an arc light, a keyhole and a molten pool, adjusting the gains of 4 images to balance the brightness of the areas of the welding gun, the arc light, the keyhole and the molten pool, and synthesizing the 4 images into an image with clear details;
step S3: outputting a predicted feature map containing a keyhole and a molten pool by using the CNN technology with the image synthesized in the step S2 as an input;
step S4: according to the prediction characteristic diagram obtained in the step S3, according to RGB color values, respectively extracting keyhole pixel points and molten pool pixel points to obtain a keyhole image and a molten pool image, respectively performing binarization, wherein the foreground is 255 and the background is 0, firstly expanding and then corroding to obtain a maximum connected domain outline, calculating an outsourcing matrix, extracting the length and width of a keyhole and the molten pool according to the outsourcing matrix, and calculating morphological characteristics of Hu invariant moment according to the maximum connected domain outline;
step S5: and identifying the fusion penetration state according to the constructed BP neural network by taking the current welding voltage, the welding current, the length and the width of the lockhole and the molten pool and morphological characteristics as input.
2. The embedded device for monitoring the welding quality of the lockhole TIG welding in real time according to claim 1, wherein the size of a PCB of the AI core board (2) is 40mmx55mm, an RK1808K wide-temperature AI chip is adopted, an internal memory is 2GB DDR4 SDRAM, the internal memory is 8GB eMMC1 channel, the embedded device is provided with an LVDS liquid crystal screen interface, 1 channel gigabit Ethernet interface, 1 channel USB3.0Host interface, 1 channel TF card interface, 1 channel audio/recording interface, 1 channel PCIE interface, 8 channels IO expansion interface, 4 channels UART interface, 3 channels SPI, 4 channels I2C, 4 channels analog acquisition ADC interface, 1 channel wireless WIFI module interface and 1 channel wireless 4G module interface;
the CMOS camera (1) has the functions of software triggering and exposure time setting, the exposure time range meets 100us-50ms, the dynamic range is larger than 60dB, and the frame rate is larger than 60 frames/s.
3. The embedded device for monitoring the welding quality of keyhole TIG welding in real time according to claim 1, wherein the power module (6) comprises a Power Management Integrated Circuit (PMIC) chip RK809, 1 configurable synchronous buck converter, 9 LDO regulators, two switches and a battery fuel meter, and the output voltage of each DC-DC converter is dynamically adjusted according to the working state of the AI core board (2) so as to maximize the efficiency of the embedded device; the power supply module (6) converts the input voltage and current of 5V/3A into various voltages to supply power to all the components of the embedded device.
4. The embedded device for lock hole TIG welding quality real-time monitoring according to claim 1, characterized in that the wireless module (5) adopts Huacheng ME909s 4G communication module to transmit the welding image thumbnail, the welding speed, the welding current voltage, the length, width and shape characteristics of the molten pool and the lock hole and the penetration state to a remote server.
5. The embedded device for monitoring the welding quality of lockhole TIG welding in real time according to claim 1, wherein the welding quality real-time detection method comprises the following steps of trial welding calibration in step S1: welding a same material according to current welding parameters, setting an exposure time sequence of a CMOS camera (1) within a range of 100us-50ms by taking 100us as a step, shooting each exposure image under the current welding parameters, manually selecting 4 exposure times respectively corresponding to a welding gun clear exposure image, a welding arc clear exposure image, a keyhole clear exposure image and a molten pool clear exposure image, setting the CMOS camera (1) to be a multi-exposure mode, and sequentially setting the exposure times to be the 4 exposure times.
6. The embedded device for monitoring the welding quality of lockhole TIG welding in real time according to claim 1, wherein the multi-exposure fusion process in step S2 of the method for detecting the welding quality in real time is as follows: and selecting areas of a welding gun, an arc, a keyhole and a molten pool as an overlapping area for calculating gain in a static frame before welding, and optimizing the gain of 4 images according to the following formula:
in the formula (I), the compound is shown in the specification,
pixel brightness g corresponding to the overlapping parts of the welding gun, arc light, keyhole and molten pool0,g1,g2,g3Respectively representing the image gain, σI,σgLet σ be the standard deviation of the overlap region error and the standard deviation of the gain, respectivelyI＝20,σg＝0.1；
Fusing the multi-exposure images according to the following formula:
in the formula (I), the compound is shown in the specification,
is the brightness, Img ', of the pixel corresponding to the unadjusted image'iAnd the brightness of the pixel points corresponding to the unadjusted image is obtained.
7. The embedded device for lock hole TIG welding quality real-time monitoring according to claim 1, wherein the CNN neural network constructed in the step S3 of the welding quality real-time detection method has an input dimension of 512x512x3 and an output dimension of 512x512x 3;
the operation of the intermediate structure of the CNN neural network is a downsampling operation, a convolution operation and an upsampling operation;
wherein the down-sampling operation: the upper layer input is subjected to common convolution of a convolution kernel of 3x3, the number of layers is consistent with that of the upper layer input, and the stepping is 2, so that first output is obtained; the upper input is subjected to average pooling, and the step number is 2 to obtain a second output; stacking the first output and the second output; the output of the layer is obtained through batch normalization and linear rectification treatment, and the length and the width of the layer are halved through downsampling operation, but the layer number is doubled;
and (3) convolution operation: the upper layer input is subjected to expansion convolution of a convolution kernel of 3x3, the number of convolution layers is consistent with that of the upper layer input, the step is 1, the output of the layer is obtained through batch normalization and linear rectification, and the length, the width and the number of layers are not changed through convolution operation;
and (3) upsampling operation: the upper layer input is subjected to deconvolution of a convolution kernel of 3x3, the number of output layers is half of the number of the upper layer input layers, the step is 2, and the output of the layer is obtained through batch normalization and linear rectification processing;
the network structure comprises an input layer 512x512x3 downsampling operation to obtain a layer 1 output 256x256x6, a layer 1 output downsampling operation to obtain a layer 2 output 128x128x12, a layer 2 output downsampling operation to obtain a layer 3 output 64x64x24, a layer 3 convolution operation to obtain a layer 4 output 64x64x24, a layer 4 convolution operation to obtain a layer 5 output 64x64x24, a layer 5 upsampling operation to obtain a layer 6 output 128x128x12, a layer 6 upsampling operation to obtain a layer 7 output 256x256x6, and a layer 7 upsampling operation to obtain an output layer 512x512x 3.
8. The embedded device for monitoring the welding quality of keyhole TIG welding in real time according to claim 7, wherein the training process of the CNN neural network is as follows:
the used training data are 5000 actual lockhole TIG welding images collected under the welding conditions of different welding speeds, welding currents, welding voltages and welding materials, the actual lockhole TIG welding images comprise states of incomplete penetration, proper penetration and excessive penetration, the labeled label images are three-channel mask images, and pixel points belonging to a background, a lockhole and a molten pool are endowed with different RGB colors respectively;
the training data were calculated as 7: and 3, dividing the training data into a training set and a verification set, simultaneously performing data enhancement on the training data in the training process, selecting data with a random probability of 25% in each training process to perform data enhancement operations including Gaussian noise, translation, scaling, mirror image, distortion and rotation, selecting the data batch size to be 8, the learning rate to be 0.001, selecting Adam by an optimizer, and selecting MSE by a loss function.
9. The embedded device for lock hole TIG welding quality real-time monitoring according to claim 7, wherein the process of extracting the length and width of the lock hole and the molten pool and the morphological characteristics according to the image processing method in the step S4 of the welding quality real-time detection method is as follows:
s41, respectively separating a keyhole image and a molten pool image according to RGB colors by the predicted characteristic diagram;
s42, respectively expanding and corroding the keyhole and the molten pool image, and then calculating a connected domain to obtain the maximum outline of the keyhole and the molten pool image;
s43, calculating the upper, lower, left and right boundaries of the maximum outline of the keyhole and the molten pool respectively to obtain an outsourcing rectangle so as to obtain the keyhole length and width characteristic Hk、WkAnd the length and width characteristics H of the molten poolp、Wp；
S44, respectively calculating the invariant moment h of the keyhole Hu according to the maximum outlines of the keyhole and the molten poolk1,hk2,…hk7Constant torque h with molten pool hup1,hp2,…hp7。
10. The embedded device for monitoring the welding quality of keyhole TIG welding in real time as claimed in claim 1, wherein the BP neural network takes the dimensional characteristics and morphological characteristics of the keyhole and the molten pool, and the current I and the voltage V measured by the welding current and voltage sensing module as input to form [ H ] Hk,Wk,Hp,Wp,hk1,hk2,…hk7,hp1,hp2,…hp7,I,V]The BP neural network comprises two hidden layers, the number of the nodes of the first hidden layer is 128, the number of the nodes of the second hidden layer is 32, a Softmax function is selected as an activation function, and the number of iterations is set as 2000; the training data used by the BP neural network are 5000 actual keyhole TIG welding images acquired under different welding conditions including different welding speeds, welding currents, welding voltages and welding materials, and the actual keyhole TIG welding images comprise states of incomplete penetration, proper penetration and excessive penetration.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110891305.6A CN113751920B (en) | 2021-08-04 | 2021-08-04 | Embedded device and method for detecting welding quality of lockhole TIG welding in real time |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110891305.6A CN113751920B (en) | 2021-08-04 | 2021-08-04 | Embedded device and method for detecting welding quality of lockhole TIG welding in real time |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113751920A true CN113751920A (en) | 2021-12-07 |
| CN113751920B CN113751920B (en) | 2022-07-22 |
Family
ID=78788554
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110891305.6A Active CN113751920B (en) | 2021-08-04 | 2021-08-04 | Embedded device and method for detecting welding quality of lockhole TIG welding in real time |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113751920B (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115106621A (en) * | 2022-05-20 | 2022-09-27 | 华南理工大学 | All-position robot deep melting K-TIG welding system and control method |
| CN117428297A (en) * | 2023-09-01 | 2024-01-23 | 广东福维德焊接技术研究院有限公司 | K-TIG welding gun short circuit, full-automatic real-time detection device for welding seam and control system |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105891215A (en) * | 2016-03-31 | 2016-08-24 | 浙江工业大学 | Welding visual detection method and device based on convolutional neural network |
| CN108596274A (en) * | 2018-05-09 | 2018-09-28 | 国网浙江省电力有限公司 | Image classification method based on convolutional neural networks |
| US20180341248A1 (en) * | 2017-05-24 | 2018-11-29 | Relativity Space, Inc. | Real-time adaptive control of additive manufacturing processes using machine learning |
| CN109719368A (en) * | 2019-01-17 | 2019-05-07 | 上海交通大学 | A kind of robot welding process multi information collection monitoring system and method |
| CN110355465A (en) * | 2019-08-21 | 2019-10-22 | 华南理工大学 | A kind of high dynamic vision system and synthetic method based on industrial camera |
| CN110956092A (en) * | 2019-11-06 | 2020-04-03 | 江苏大学 | Intelligent metallographic detection and rating method and system based on deep learning |
| CN111539533A (en) * | 2020-04-20 | 2020-08-14 | 上海工程技术大学 | Welding penetration quantitative evaluation method based on extreme learning machine and small hole characteristics |
| CN112365466A (en) * | 2020-11-10 | 2021-02-12 | 嘉兴凡视智能科技有限公司 | Welding spot appearance detection method based on AI |
-
2021
- 2021-08-04 CN CN202110891305.6A patent/CN113751920B/en active Active
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105891215A (en) * | 2016-03-31 | 2016-08-24 | 浙江工业大学 | Welding visual detection method and device based on convolutional neural network |
| US20180341248A1 (en) * | 2017-05-24 | 2018-11-29 | Relativity Space, Inc. | Real-time adaptive control of additive manufacturing processes using machine learning |
| CN108596274A (en) * | 2018-05-09 | 2018-09-28 | 国网浙江省电力有限公司 | Image classification method based on convolutional neural networks |
| CN109719368A (en) * | 2019-01-17 | 2019-05-07 | 上海交通大学 | A kind of robot welding process multi information collection monitoring system and method |
| CN110355465A (en) * | 2019-08-21 | 2019-10-22 | 华南理工大学 | A kind of high dynamic vision system and synthetic method based on industrial camera |
| CN110956092A (en) * | 2019-11-06 | 2020-04-03 | 江苏大学 | Intelligent metallographic detection and rating method and system based on deep learning |
| CN111539533A (en) * | 2020-04-20 | 2020-08-14 | 上海工程技术大学 | Welding penetration quantitative evaluation method based on extreme learning machine and small hole characteristics |
| CN112365466A (en) * | 2020-11-10 | 2021-02-12 | 嘉兴凡视智能科技有限公司 | Welding spot appearance detection method based on AI |
Non-Patent Citations (2)
| Title |
|---|
| B. ZHANG ET AL.: "Prediction of keyhole TIG weld penetration based on high-dynamic", 《JOURNAL OF MANUFACTURING PROCESSES》 * |
| 梁斌等: "深熔K-TIG焊接过程的声信号采集及特征分析", 《电焊机》 * |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115106621A (en) * | 2022-05-20 | 2022-09-27 | 华南理工大学 | All-position robot deep melting K-TIG welding system and control method |
| CN115106621B (en) * | 2022-05-20 | 2023-05-23 | 华南理工大学 | Full-position robot deep-melting K-TIG welding system and control method |
| CN117428297A (en) * | 2023-09-01 | 2024-01-23 | 广东福维德焊接技术研究院有限公司 | K-TIG welding gun short circuit, full-automatic real-time detection device for welding seam and control system |
| CN117428297B (en) * | 2023-09-01 | 2024-04-16 | 广东福维德焊接技术研究院有限公司 | K-TIG welding gun short circuit, full-automatic real-time detection device for welding seam and control system |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113751920B (en) | 2022-07-22 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN113751920B (en) | Embedded device and method for detecting welding quality of lockhole TIG welding in real time | |
| Zhu et al. | Modified densenet for automatic fabric defect detection with edge computing for minimizing latency | |
| CN109900706B (en) | Weld joint based on deep learning and weld joint defect detection method | |
| WO2021248554A1 (en) | High-speed and high-precision burr detection method and detection system for lithium ion battery electrode sheet | |
| CN107808131B (en) | Dynamic gesture recognition method based on dual-channel deep convolutional neural network | |
| CN111223088A (en) | Casting surface defect identification method based on deep convolutional neural network | |
| CN105184779B (en) | One kind is based on the pyramidal vehicle multiscale tracing method of swift nature | |
| CN111914698B (en) | Human body segmentation method, segmentation system, electronic equipment and storage medium in image | |
| CN108765416A (en) | PCB surface defect inspection method and device based on fast geometric alignment | |
| CN112304960B (en) | High-resolution image object surface defect detection method based on deep learning | |
| CN113724216A (en) | Method and system for detecting wave crest welding spot defects | |
| CN112907519A (en) | Metal curved surface defect analysis system and method based on deep learning | |
| WO2024037408A1 (en) | Underground coal mine pedestrian detection method based on image fusion and feature enhancement | |
| CN112419237A (en) | Automobile clutch master cylinder groove surface defect detection method based on deep learning | |
| CN113554587B (en) | Deep learning-based molten pool image geometric feature extraction method and system | |
| CN108067714A (en) | A kind of thin-walled circumferential weld termination quality on-line monitoring and defect positioning system and method | |
| CN112819756B (en) | PCB surface defect detection device and method | |
| WO2023207064A1 (en) | Maskrcnn water seepage detection method and system based on weak light compensation | |
| CN113409355A (en) | Moving target identification system and method based on FPGA | |
| CN111368637B (en) | Transfer robot target identification method based on multi-mask convolutional neural network | |
| Zhu et al. | Advanced driver assistance system based on machine vision | |
| CN116309398A (en) | Printed circuit board small target defect detection method based on multi-channel feature fusion learning | |
| CN116469164A (en) | Human gesture recognition man-machine interaction method and system based on deep learning | |
| CN110147818B (en) | Sparse representation-based laser welding forming defect prediction classification method | |
| He et al. | Research on Weld Recognition Method Based on Mask R-CNN |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| OL01 | Intention to license declared | ||
| OL01 | Intention to license declared |