[go: up one dir, main page]
More Web Proxy on the site http://driver.im/

CN109886298B - Weld quality detection method based on convolutional neural network - Google Patents

Weld quality detection method based on convolutional neural network Download PDF

Info

Publication number
CN109886298B
CN109886298B CN201910041065.3A CN201910041065A CN109886298B CN 109886298 B CN109886298 B CN 109886298B CN 201910041065 A CN201910041065 A CN 201910041065A CN 109886298 B CN109886298 B CN 109886298B
Authority
CN
China
Prior art keywords
image
layer
welding
weld
neural network
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.)
Active
Application number
CN201910041065.3A
Other languages
Chinese (zh)
Other versions
CN109886298A (en
Inventor
陆虎
余超杰
姚棋
刘赛雄
朱玉全
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Chengdu Rongsheng Technology Co ltd
Shenzhen Wanzhida Technology Co ltd
Original Assignee
Chengdu Rongsheng Technology Co ltd
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Chengdu Rongsheng Technology Co ltd filed Critical Chengdu Rongsheng Technology Co ltd
Priority to CN201910041065.3A priority Critical patent/CN109886298B/en
Publication of CN109886298A publication Critical patent/CN109886298A/en
Application granted granted Critical
Publication of CN109886298B publication Critical patent/CN109886298B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • YGENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02PCLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
    • Y02P90/00Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
    • Y02P90/30Computing systems specially adapted for manufacturing

Landscapes

  • Image Analysis (AREA)

Abstract

The invention discloses a weld quality detection method based on a convolutional neural network. Image acquisition: collecting a welding point image, and recording a time mark collecting function and a weldment information recording function; secondly, weld quality analysis: and 1, capturing and positioning a welding seam area by adopting a convolutional neural network, positioning and extracting a welding seam detection area through a rectangular frame, preprocessing the detected welding seam area image, and sending the preprocessing result into the step 2 to detect and analyze the welding seam quality. And step 2, performing quality detection on the welding point area, judging whether the welding line to be detected is qualified, amplifying the captured welding point area by using cubic spline interpolation, performing welding line quality detection on the amplified welding area through a convolutional neural network, performing detection model training after preprocessing the welding point sample and non-welding point sample data, and performing welding line quality analysis detection by using a detection model. And step 3, carrying out corresponding processing on the detection result.

Description

Weld quality detection method based on convolutional neural network
Technical Field
The invention belongs to the field of machine vision, in particular to the field of industrial image detection and identification, and particularly relates to a weld surface image detection method based on a convolutional neural network. And detecting the quality of the welding seam through detecting and identifying the welding seam surface image.
Background
On the production line of manual welding, the phenomena of false welding, missing welding, overwelding and the like are often caused due to the immature welding process, and missing welding is more easily caused for more complex workpieces, especially for the condition of multiple welding spots on complex welding surfaces. At present, the quality of a welding spot is mostly detected manually, so that the quality is greatly dependent on the technology and experience of a detection worker, a plurality of unstable factors can be generated, the workload of the worker is increased, and the working efficiency is low. The traditional image detection and identification method is used for correcting images of multiple welding spots on a complex welding surface by correcting the images of each angle, detecting the images by an image positioning and template matching method, and is not strong in stability and easy to be interfered by the environment, so that missed detection or false detection is caused; or the sliding window on the image is used for extracting the characteristics in the window and carrying out classification detection by combining with the classifier, the sliding window on the image can cause the increase of the calculated amount, and the manually extracted image characteristics and the parameter adjustment of the classifier can influence the detection and identification precision of the welding spot image.
Related patents such as patent application 201710818297.6, which are related to the prior art, are based on a deep learning welding spot quality detection method, and only the processed target area is processed. How to capture and position the detection area through a convolutional neural network and finish the quality detection of the welding line is not known at present.
The invention provides a weld quality detection method based on a convolutional neural network. And capturing and positioning a welding line area to be detected, and detecting the quality of the area.
Disclosure of Invention
Aiming at the defects of the existing welding seam quality detection technology, the invention provides a welding seam quality detection screening method based on a convolutional neural network. And (3) carrying out quality analysis on the welding seam by collecting the image of the surface of the welding seam, and screening welding pieces with unqualified quality according to analysis results.
In order to achieve the above purpose, the technical scheme provided by the invention is a weld surface quality detection method based on a convolutional neural network, which comprises the following steps: and (5) image acquisition, weld quality analysis and detection result processing.
Image acquisition: the high-resolution industrial camera is adopted to realize the functions of scanning and recording the welding point images, realizing the functions of collecting and transmitting the welding point images to the quality analysis module, and recording the collecting time marking function and the weldment information recording function.
Secondly, weld quality analysis: the weld quality analysis comprises the following steps:
step 1, performing weld joint region positioning capturing on an image to be detected: capturing and positioning a welding seam area by adopting a convolutional neural network, positioning and extracting a welding seam detection area by a rectangular frame, preprocessing the detected welding seam area image, and sending the preprocessing result to a quality detection analysis module in the step 2 for detecting and analyzing the welding seam quality.
And 2, detecting the quality of the captured welding point area, and judging whether the welding line to be detected is qualified or not. To ensure the sharpness of the weld area, the captured weld area is scaled up by cubic spline interpolation. And amplifying the welding area detected previously, and detecting the weld quality of the area through a convolutional neural network. The test model training was performed after preprocessing a large number of weld and non-weld sample data (12500 samples for quality test models for the normal and abnormal weld samples in this example). Specifically:
and 2.1, training a convolutional neural network. Under the tensorsurface framework of the Ubuntu system, the pre-processed 12500 normal weld images and 12500 Zhang Yichang weld image data are divided into a training set, a training set label, a testing set and a testing set label. A back propagation learning algorithm and a random gradient descent method are adopted. And carrying out back propagation iteration updating on the weight of each layer of the convolutional neural network according to the magnitude of the loss value of the forward propagation. Until the loss value of the model tends to converge, it indicates that model training is complete.
And 2.2, extracting the characteristics of the weld image. And inputting each image in the data set into the convolutional neural network in the previous step, extracting the characteristics of the weld image under the model according to the trained detection model, and obtaining the characteristics of the image by adopting full convolution processing on the penultimate layer of the model.
And 2.3, recognizing a welding line image. Giving a weld image to be classified, inputting the weld image into a trained deep learning model to obtain the characteristics of the image under a convolutional neural network, and adopting a classifier to detect the quality of the image at the last layer of the network model.
Thirdly, processing detection results: the detection result processing module comprises: the welding part alarm screening device comprises a data storage module, a data display module, a welding part alarm screening module and an abnormal data storage module.
The invention has the beneficial effects that:
the method has the advantages that the convolutional neural network performs welding quality analysis on the welding seam image, complicated detection steps in the existing method are avoided, and the welding seam area can be automatically positioned and captured. In addition, the display module can monitor quality detection in real time and alarm reminding.
Drawings
FIG. 1 is a flow chart of the method design of the present invention.
FIG. 2 is a block diagram of a neural network for weld area location capture in accordance with the present invention.
Fig. 3 is a diagram of a network structure for detecting a target area to be selected.
FIG. 4 is a flow chart of weld quality detection in accordance with the present invention.
FIG. 5 is a diagram of a neural network for weld quality detection according to the present invention.
Detailed Description
The invention is further described below with reference to the accompanying drawings.
The whole implementation flow of the invention is as shown in figure 1, the welding seam image is collected, the collected image information is stored, the welding seam to-be-detected area is positioned and captured on the collected image, and the quality detection is carried out on the captured detection area.
The following processing is carried out on the detection result:
1. updating the image information and adding the quality detection result to the image information item.
2. And saving the image information of the welding line of the abnormal result.
3. And displaying the weld image quality detection result information.
4. And carrying out alarm processing on the detected abnormal weld image analysis result.
5. And screening the welding parts according to the abnormal weld information analysis result.
The specific implementation is as follows:
image acquisition
The welding point image acquisition module is realized by adopting the existing high-resolution camera, has the functions of scanning and recording welding point images, realizes the functions of acquiring and transmitting the welding point images to the quality analysis module, and has the functions of recording acquisition time marks and weldment information recording.
And the image collector sends the collected welding point information to the quality analysis module after completing the collection, wherein the welding point information comprises a welding point image, image collection time and a weldment label mark.
And (3) folding the welding point images: such as surface porosity, undercut, flash, burn-through, weld surface cracking, weld dimensional deviations, etc. of the weld. Before image acquisition, the spatter and dirt around the weld should be cleaned.
Secondly, weld quality analysis
The weld quality analysis comprises the following steps:
step 1, carrying out positioning capturing on a detection welding line area on an image to be detected:
capturing and positioning a welding seam area by adopting a convolutional neural network, positioning and extracting a welding seam detection area by a rectangular frame, preprocessing the detected welding seam area image, and sending the result to a quality detection analysis module for detecting and analyzing the welding seam quality.
The design of the convolutional neural network shown in fig. 2 includes the steps of:
convolution layer: to be able to detect the localization to the target area, a feature map of the image is first extracted using a set of underlying convolutional layer + active layer + pooling layer. The feature map is shared for subsequent RPN (Region Proposal Networks) layers and full connection layers. All convolution layers are: the convolution kernel size is 3 and the image frame is 1. All pooling layers are: the convolution kernel size is 2 and the convolution kernel movement step size is 2.
Fig. 3 shows RPN (Region Proposal Networks) layers, the RPN network being used to generate alternative target detection zones. The layer judges that the region set belongs to a target foreground or background through softmax, and then corrects the region set by frame regression to obtain an accurate candidate target region.
Generating a region set: the super pixel is taken as a unit, the frame selection area is each super pixel area obtained by dividing the super pixel, and the super pixel critical area is obtained by the super pixel area and the super pixel critical area, so that the calculation amount can be reduced.
Rectangular frame adjustment standard: diffusion or contraction is performed according to the critical matrix of super pixels. The size of the rectangular frame to be selected is the largest rectangular frame occupied by the generation super pixel or super pixel group.
Frame sliding: image traversal based on breadth-first. And according to the traversing path, sliding the frame to select the frame, wherein each sliding ensures that the rectangular frame area is the largest rectangular area covered by the current super pixel.
And (3) frame adjustment: based on the contiguous superpixel maximum similarity max (s i ) Adjustment of superpixel merging
Combining rules: calculating the similarity of super pixels: s is(s) i =d(x i ,y i ) Wherein s is i Covering superpixel x for rectangle i Super pixel y next to i Is used for covering super pixel x in a rectangle i Super pixel y next to i Is represented by the euclidean distance. And merging the super pixels with the maximum similarity in the critical super pixels to form a new super pixel area. The rectangular border is the largest rectangular area covered by the new super pixel area.
Frame quality evaluation IOU (Interaction-over-Union) value calculation: the intersection between the salient region detection result and the standard result label is adopted as the value of the IOU on the union between the salient region detection result and the standard result label.
Target area judgment: and calculating whether the detected region is a target region or not by using the feature map of the detected region, and simultaneously, adjusting the frame region again to obtain the final accurate position of the detection frame.
The process of detecting the quality of the captured region to be welded and judging whether the welded joint is qualified is shown in fig. 4. And inputting the image to be detected as a weld joint image, extracting the characteristics of the image through a trained deep convolutional neural network model, classifying according to the quality class of the weld joint, and finally outputting a classification result.
In order to ensure the definition of the welding seam area, the captured welding seam area is amplified by cubic spline interpolation, and after the previously detected suspected welding seam area is amplified, the welding seam quality detection is carried out on the area through a convolution neural network. Training was performed by preprocessing a large number of normal bead samples and abnormal bead samples (12500 samples of the normal bead samples and the abnormal bead samples trained in this example were used for the quality inspection model).
(1) Design module of convolutional neural network
Fig. 5 is a schematic diagram of a convolutional neural network model according to the present invention, which comprises an input layer, a hidden layer, and an output layer:
A. input layer
The input layer captures a weld image of the localized area for the data weld.
B. Hidden layer
Convolution layer: and sequentially selecting convolution layers with convolution kernels of 3 x 3 and 1 x 1, wherein the number of channels of each layer is the same as that of the input layer.
Activation function:
Figure GDA0003941044720000051
where x represents the feature map to be activated in each layer.
Pooling layer: and using a convolution kernel of 2 x 2, and carrying out maximum pooling operation on the output after the convolution operation after the convolution layer.
C. Output layer
The output layer is connected with the last full convolution layer of the hidden layer, the output dimension is two, and each dimension represents the probability of a normal welding line image and an abnormal welding line image.
The method comprises 6 groups of convolution layers and 1 group of full connection layers; each of the first 4 sets of convolutional layers includes a convolutional layer with a convolutional kernel 3*3, an active layer, and a pooling layer. The latter two groups are convolution layers with a convolution kernel of 1*1. And finally, outputting a welding seam quality detection result by the full-connection layer.
(2) Training of convolutional neural networks
Under the tensorsurface framework of the Ubuntu system, the preprocessed 12500 normal weld images and the processed image data of the 12500 Zhang Yichang weld are divided into a training set, a training set label, a testing set and a testing set label. A back propagation learning algorithm and a random gradient descent method are adopted. And carrying out back propagation iteration updating on the weight of each layer of the convolutional neural network according to the magnitude of the loss value of the forward propagation. Until the loss value of the model tends to converge, it indicates that model training is complete.
(3) Feature extraction of weld images
And inputting each image in the data set into the convolutional neural network in the previous step, and extracting the characteristics of the weld image under the model according to the trained detection model, wherein the characteristics are obtained by the penultimate full convolutional layer.
(4) Identification of weld images
Giving a welding spot image to be classified, inputting the welding spot image to the convolutional neural network model trained by the method, obtaining the characteristics of the image under the convolutional neural network model, and detecting the quality of the image in the last layer.
Third, result processing
The data result processing comprises the following steps: and (5) data storage, data display, weldment alarm screening and abnormal data storage.
And (3) data storage: the storage module is used for storing the image data acquired by the image acquisition module and the output data of the data analysis module, and each piece of data comprises: and (5) weld quality analysis results, image acquisition time and weld mark number.
And (3) data display: and displaying the trend of the welding seam detection result in the image acquisition time axis by reading the data of the storage module. And when unqualified welding spot data occurs, displaying detection abnormality alarm.
And (3) abnormal data storage: storing, by an abnormal data storage module, abnormal pad data, each piece of data including: and numbering weldments where the defective weld joints are located, collecting images, and detecting quality.
Fourth, weldment screening alarm
When the quality detection result is lower than the normal threshold value, the following steps are executed:
1. storing weldment information and detection results to an abnormal data storage module
2. And sending an abnormal signal to an alarm system, triggering an alarm, displaying abnormal data by a display module and giving an alarm.
3. And marking defective weldments according to the detected abnormal result information.
4. And screening according to abnormal data defective weldments.
The above list of detailed descriptions is only specific to practical embodiments of the present invention, and they are not intended to limit the scope of the present invention, and all equivalent embodiments or modifications that do not depart from the spirit of the present invention should be included in the scope of the present invention.

Claims (5)

1. The welding seam quality detection method based on the convolutional neural network is characterized by comprising the following steps of:
step one, image acquisition: collecting a welding point image, and recording the collecting time and weldment information;
step two, weld quality analysis: the method comprises the following steps:
step 1, performing weld joint region positioning capturing on an image to be detected: capturing and positioning a welding seam area by adopting a convolutional neural network, positioning and extracting a welding seam detection area by a rectangular frame, preprocessing a detected welding seam area image, and sending a preprocessing result to a quality detection analysis module in the step 2 for detecting and analyzing the welding seam quality;
step 2, quality detection is carried out on the captured welding point area, and whether the welding line to be detected is qualified or not is judged; firstly, in order to ensure the definition of a welding joint region, amplifying the captured welding joint region by using cubic spline interpolation, then amplifying the detected welding region, detecting the quality of the welding joint of the region through a convolutional neural network, preprocessing a large number of welding joint samples and non-welding joint sample data, training a detection model, and analyzing and detecting the quality of the welding joint by using the trained model;
thirdly, carrying out corresponding treatment on the detection result;
in the step 2, training 12500 normal weld samples and 12500 abnormal weld samples for a quality detection model;
the specific implementation of the step 2 comprises the following steps:
step 2.1, training a convolutional neural network model: dividing the pre-processed 12500 normal weld images and 12500 Zhang Yichang weld image data into a training set, a training set label, a test set and a test set label, adopting a back propagation learning algorithm and a random gradient descent method, and carrying out back propagation iteration updating on the weight of each layer of the convolutional neural network according to the magnitude of a loss value of forward propagation until the loss value of a model tends to converge, so that model training is completed;
step 2.2, extracting characteristics of the weld image: inputting each image in the data set into the convolutional neural network in the previous step, extracting the characteristics of the weld image under the model according to the trained convolutional neural network detection model, and obtaining the characteristics of the image by adopting full convolution processing on the penultimate layer of the model;
step 2.3, recognizing a welding line image: giving a weld image to be detected, inputting the weld image into a trained deep learning model to obtain the characteristics of the image under a convolutional neural network, and detecting the quality of the image by using a classifier at the last layer of the network model;
in step 2.1, the design method of the convolutional neural network model includes:
designing a convolution layer: firstly, extracting a characteristic map of an image by using a group of basic convolution layer, an activation layer and a pooling layer, wherein the characteristic map is shared for the subsequent RPN layer and the full connection layer; all convolution layers are: the convolution kernel size is 3, and the image frame is 1; all pooling layers are: the size of the convolution kernel is 2, and the moving step length of the convolution kernel is 2;
designing an RPN layer: the RPN layer is used for generating an alternative target detection area, judging that the area set belongs to a target foreground or background through softmax, and correcting the area set by frame regression to obtain an accurate alternative target area;
designing a region set: the generation of the region set is to take super pixels as a unit, the frame selection region is each super pixel region and super pixel critical region obtained by dividing super pixels, and the region set is obtained by the super pixel region and the super pixel critical region;
designing a rectangular frame adjusting standard: performing diffusion or contraction according to the critical matrix of the super pixels, wherein the size of the rectangular frame to be selected is the maximum rectangular frame occupied by the super pixels or the super pixel groups;
design frame sliding: based on breadth-first image traversal, sliding frames according to traversal paths, wherein each sliding ensures that a rectangular frame area is the largest rectangular area covered by the current super pixel;
and (3) adjusting a design frame: based on the contiguous superpixel maximum similarity max (s i ) Performing super-pixel merging adjustment;
designing a merging rule: calculating the similarity of super pixels: s is(s) i =d(x i ,y i ) Wherein s is i Covering superpixel x for rectangle i Super pixel y next to i Is used for covering super pixel x in a rectangle i Super pixel y next to i Is a Euclidean distance representation of (2); merging super pixels with the maximum similarity in the critical super pixels to form a new super pixel area, wherein the rectangular frame is the largest rectangular area covered by the new super pixel area;
designing a frame quality evaluation (IOU) value calculation method: the intersection between the significant region detection result and the standard result label is adopted as the value of the IOU on the union between the significant region detection result and the standard result label;
the design target area judging method comprises the following steps: and calculating whether the detected region is a target region or not by using the feature map of the detected region, and simultaneously, adjusting the frame region again to obtain the final accurate position of the detection frame.
2. The weld quality detection method based on the convolutional neural network according to claim 1, wherein the training of the convolutional neural network model is realized under a tensorflow framework of a Ubuntu system.
3. The method for detecting the quality of a weld joint based on a convolutional neural network according to claim 1, wherein the designed convolutional neural network model comprises an input layer, a hidden layer and an output layer, wherein:
input layer: the input layer captures a weld image of the positioning area for the data weld;
hidden layer: comprises a convolution layer and a pooling layer; convolution layer: sequentially selecting convolution layers with convolution kernels of 3 x 3 and 1 x 1, wherein the number of channels of each layer is the same as that of an input layer; the activation function is set to:
Figure FDA0003941044710000031
wherein x represents a feature map to be activated in each layer; pooling layer: using a convolution kernel of 2 x 2, and performing maximum pooling operation on the output after the convolution operation after the convolution layer;
output layer: the output layer is connected with the last full convolution layer of the hidden layer, the output dimension is 2, and each dimension represents the probability of a normal weld image and an abnormal weld image; the method comprises 6 groups of convolution layers and 1 group of full connection layers; each of the first 4 groups of convolution layers comprises a convolution layer with a convolution kernel of 3*3, an activation layer and a pooling layer, the second two groups of convolution layers are convolution layers with a convolution kernel of 1*1, and the last full-connection layer is used for outputting a weld quality detection result.
4. The method for detecting the quality of a weld seam based on a convolutional neural network according to claim 1, wherein the processing of the detection result in the third step comprises: updating image information, and/or adding quality detection results to image information items, and/or storing welding seam image information of abnormal results, and/or displaying welding seam image quality detection result information, and/or carrying out alarm processing on detected abnormal welding seam image analysis results, and/or screening and alarming welding parts according to the abnormal welding seam information analysis results.
5. The method for detecting the quality of a welding seam based on a convolutional neural network according to claim 4, wherein the screening and alarming of the welding seam is performed when the detection result is lower than a normal threshold value, by performing the following steps:
1) Storing the weldment information and the detection result to an abnormal data storage module;
2) Sending an abnormal signal to an alarm system, triggering an alarm, and displaying abnormal data and alarming by a display module;
3) Marking defective weldments according to the detected abnormal result information;
4) Screening the defective weldments according to the abnormal data.
CN201910041065.3A 2019-01-16 2019-01-16 Weld quality detection method based on convolutional neural network Active CN109886298B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201910041065.3A CN109886298B (en) 2019-01-16 2019-01-16 Weld quality detection method based on convolutional neural network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201910041065.3A CN109886298B (en) 2019-01-16 2019-01-16 Weld quality detection method based on convolutional neural network

Publications (2)

Publication Number Publication Date
CN109886298A CN109886298A (en) 2019-06-14
CN109886298B true CN109886298B (en) 2023-06-16

Family

ID=66926043

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201910041065.3A Active CN109886298B (en) 2019-01-16 2019-01-16 Weld quality detection method based on convolutional neural network

Country Status (1)

Country Link
CN (1) CN109886298B (en)

Families Citing this family (19)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110570400B (en) * 2019-08-19 2022-11-11 河北极目楚天微电子科技有限公司 Information processing method and device for chip 3D packaging detection
CN111037549B (en) * 2019-11-29 2022-09-09 重庆顺泰铁塔制造有限公司 Welding track processing method and system based on 3D scanning and TensorFlow algorithm
CN111421231A (en) * 2020-04-07 2020-07-17 湖南汽车工程职业学院 Omnibearing laser welding production line and welding method thereof
DE102020204522A1 (en) * 2020-04-08 2021-10-14 Robert Bosch Gesellschaft mit beschränkter Haftung Method for optimizing welding parameters for a welding controller, method for providing a trained machine learning algorithm and welding controller
CN113808065A (en) * 2020-05-28 2021-12-17 青岛海尔工业智能研究院有限公司 Weld joint detection method and device, industrial robot and storage medium
CN112171057B (en) * 2020-09-10 2022-04-08 五邑大学 Quality detection method and device based on laser welding and storage medium
CN112183957A (en) * 2020-09-10 2021-01-05 五邑大学 Welding quality detection method and device and storage medium
CN111882557B (en) * 2020-09-28 2021-01-05 成都睿沿科技有限公司 Welding defect detection method and device, electronic equipment and storage medium
CN112285114A (en) * 2020-09-29 2021-01-29 华南理工大学 Enameled wire spot welding quality detection system and method based on machine vision
CN113325068B (en) * 2021-04-29 2024-02-02 河南工业大学 Weld quality detection method and detection system based on fuzzy control
CN113432644A (en) * 2021-06-16 2021-09-24 苏州艾美睿智能系统有限公司 Unmanned carrier abnormity detection system and detection method
CN113421304B (en) * 2021-06-21 2024-05-28 沈阳派得林科技有限责任公司 Intelligent positioning method for welding bead area of industrial ray film image
CN113256620B (en) * 2021-06-25 2021-09-28 南京思飞捷软件科技有限公司 Vehicle body welding quality information judging method based on difference convolution neural network
CN113478483B (en) * 2021-07-02 2022-12-09 普瑞特机械制造股份有限公司 Mobile robot welding method and system based on stainless steel storage tank
CN113658132B (en) * 2021-08-16 2022-08-19 沭阳九鼎钢铁有限公司 Computer vision-based structural part weld joint detection method
CN116038112A (en) * 2022-12-06 2023-05-02 西南石油大学 Laser tracking large-scale curved plate fillet welding system and method
DE102023201384A1 (en) 2023-02-17 2024-08-22 Robert Bosch Gesellschaft mit beschränkter Haftung Method and system for qualifying a welded joint between at least two components
KR20240155529A (en) * 2023-04-20 2024-10-29 현대모비스 주식회사 System and method for checking quality of laser welding
CN117047274B (en) * 2023-09-05 2024-03-19 浙江人驰汽车配件有限公司 Internal bracing type double-welding tensioning wheel and intelligent processing method thereof

Family Cites Families (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105891215B (en) * 2016-03-31 2019-01-29 浙江工业大学 Welding visible detection method and device based on convolutional neural networks
CN109001211A (en) * 2018-06-08 2018-12-14 苏州赛克安信息技术有限公司 Welds seam for long distance pipeline detection system and method based on convolutional neural networks
CN109142547B (en) * 2018-08-08 2021-02-23 广东省智能制造研究所 Acoustic online nondestructive testing method based on convolutional neural network

Also Published As

Publication number Publication date
CN109886298A (en) 2019-06-14

Similar Documents

Publication Publication Date Title
CN109886298B (en) Weld quality detection method based on convolutional neural network
CN113362326B (en) Method and device for detecting defects of welding spots of battery
CN111080622B (en) Neural network training method, workpiece surface defect classification and detection method and device
CN111986190B (en) Printed matter defect detection method and device based on artifact elimination
CN111815572B (en) Method for detecting welding quality of lithium battery based on convolutional neural network
CN106952250A (en) A kind of metal plate and belt detection method of surface flaw and device based on Faster R CNN networks
CN106934800A (en) A kind of metal plate and belt detection method of surface flaw and device based on YOLO9000 networks
CN109859177A (en) Industrial x-ray image assessment method and device based on deep learning
CN113393426B (en) Steel rolling plate surface defect detection method
CN109308489A (en) Method for detecting welding quality of element arc welding
US20150051860A1 (en) Automatic optical appearance inspection by line scan apparatus
CN111415339B (en) Image defect detection method for complex texture industrial product
CN114240944A (en) Welding defect detection method based on point cloud information
CN115035092A (en) Image-based bottle detection method, device, equipment and storage medium
CN116228651A (en) Cloth defect detection method, system, equipment and medium
CN112200790B (en) Cloth defect detection method, device and medium
CN110728269B (en) High-speed rail contact net support pole number plate identification method based on C2 detection data
CN117455917B (en) Establishment of false alarm library of etched lead frame and false alarm on-line judging and screening method
CN114235810B (en) Cloth flaw automatic detection method based on area array industrial camera
CN114120318B (en) Dark field image target point accurate extraction method based on integrated decision tree
CN116958073A (en) Small sample steel defect detection method based on attention feature pyramid mechanism
CN116912189A (en) Welding spot detection method and system based on deep learning
CN110147818B (en) Sparse representation-based laser welding forming defect prediction classification method
CN116008304A (en) Weld joint detection method suitable for pipeline interior
TW202222030A (en) Method, computer program, and computer readable medium of using electroluminescence images to identify defect of photovoltaic module based on deep learning technology

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
TA01 Transfer of patent application right
TA01 Transfer of patent application right

Effective date of registration: 20230518

Address after: Room 12, 3rd Floor, Building 3, No. 2 Xingsheng West Road, Jinniu High tech Industrial Park, Chengdu City, Sichuan Province, 610037

Applicant after: Chengdu Rongsheng Technology Co.,Ltd.

Address before: Room 1002-1003, Building A, Zhiyun Industrial Park, No. 13 Huaxing Road, Henglang Community, Dalang Street, Longhua District, Shenzhen City, Guangdong Province, 518109

Applicant before: Shenzhen Wanzhida Technology Co.,Ltd.

Effective date of registration: 20230518

Address after: Room 1002-1003, Building A, Zhiyun Industrial Park, No. 13 Huaxing Road, Henglang Community, Dalang Street, Longhua District, Shenzhen City, Guangdong Province, 518109

Applicant after: Shenzhen Wanzhida Technology Co.,Ltd.

Address before: Zhenjiang City, Jiangsu Province, 212013 Jingkou District Road No. 301

Applicant before: JIANGSU University

GR01 Patent grant
GR01 Patent grant